4 # Copyright: 08/04/2012: v1.0.2 Glyn Astill <glyn@8kb.co.uk>
5 # Requires: Perl 5.10.1+, Data::UUID, File::Slurp
6 # PostgreSQL 9.0+ Slony-I 1.2+ / 2.0+
8 # This script is a command-line utility to manage switchover and failover
9 # of replication sets in Slony-I clusters.
11 # This script is free software: you can redistribute it and/or modify
12 # it under the terms of the GNU General Public License as published by
13 # the Free Software Foundation, either version 3 of the License, or
14 # (at your option) any later version.
16 # This script is distributed in the hope that it will be useful,
17 # but WITHOUT ANY WARRANTY; without even the implied warranty of
18 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 # GNU General Public License for more details.
21 # You should have received a copy of the GNU General Public License
22 # along with this script. If not, see <http://www.gnu.org/licenses/>.
27 use Getopt::Long qw/GetOptions/;
31 use sigtrap 'handler' => \&sigExit, 'HUP', 'INT','ABRT','QUIT','TERM';
32 use Time::HiRes qw/usleep gettimeofday/;
33 use POSIX qw/strftime/;
34 use Config qw/%Config/;
36 use constant false => 0;
37 use constant true => 1;
39 my $g_script_version = '1.0.3';
41 my $g_pidfile = '/var/run/slony_failover.pid';
42 my $g_pid_written = false;
44 my $g_prefix = '/tmp/slony_failovers';
45 my $g_separate_working = true;
46 my $g_log_prefix = '%t';
47 my $g_log_to_db = false;
54 my $g_use_try_blocks = false;
55 my $g_lockset_method = 'multiple';
56 my $g_logfile = 'failover.log';
58 my $g_silence_notice = false;
69 my $g_dbuser = 'slony';
71 my $g_available_node_count;
72 my $g_critical_node_count;
73 my $g_subs_follow_origin = false;
74 my $g_use_comment_aliases = false;
75 my @g_cluster; # Array refs of node info. In hindsight this should have been a hash, should be fairly simple to switch.
81 my $g_failover_method = 'old';
82 my $g_resubscribe_method = 'subscribe';
83 my $g_failover = false;
84 my $g_fail_subonly = false;
85 my $g_drop_failed = false;
86 my $g_autofailover = false;
87 my $g_autofailover_poll_interval = 500;
88 my $g_autofailover_retry = 2;
89 my $g_autofailover_retry_sleep = 1000;
90 my $g_autofailover_provs = false;
91 my $g_autofailover_config_any = true;
92 my $g_autofailover_perspective_sleep = 20000;
93 my $g_autofailover_majority_only = false;
94 my $g_autofailover_is_quorum = false;
96 my %g_unresponsive_subonly;
100 # Hash containing messages used by lookupMsg()
103 'usage' => q{-h <host> -p <port> -db <database> -cl <cluster name> -u <username> -P <password> -f <config file> (Password option not recommended; use pgpass instead)},
104 'title' => q{Slony-I failover script version $1},
105 'cluster_fixed' => q{Aborting failover action: all origin/provider nodes now responsive},
106 'cluster_failed' => q{Found $1 failed nodes, sleeping for $2ms before retry $3 of $4},
107 'load_cluster' => q{Getting a list of database nodes...},
108 'load_cluster_fail' => q{Unable to read cluster configuration $1},
109 'load_cluster_success' => q{I Loaded Slony-I v$1 cluster "$2" with $3 nodes read from node at $4:$5/$6},
110 'lag_detail' => q{Current node lag information from configuration node:},
111 'script_settings' => q{Using $1 batches of lock set, $2 FAILOVER and $3},
112 'generated_script' => q{Generated script "$1"},
113 'autofailover_init' => q{Entering autofailover mode},
114 'autofailover_init_cnf' => q{Slony configuration will be read from $1 node},
115 'autofailover_init_pol' => q{Polling every $1ms},
116 'autofailover_init_ret' => q{Failed nodes will be retried $1 times with $2ms sleep},
117 'autofailover_init_set' => q{Failed forwarding providers $1 be failed over},
118 'autofailover_load_cluster' => q{$1 Slony-I v$2 cluster "$3" with $4 nodes read from node $5},
119 'autofailover_proceed' => q{Proceeding with failover:},
120 'autofailover_detail' => q{Failed node: $1, Backup node: $2},
121 'autofailover_halt' => q{Unable to perform any failover for $1 failed nodes},
122 'autofailover_check_sub' => q{Checking subscriber node $1},
123 'autofailover_check_sub_fail' => q{Unable to check subscriber node $1},
124 'autofailover_promote_find' => q{Finding most up to date subscriber to all sets ($1) on unresponsive node $2},
125 'autofailover_promote_found' => q{Using previously found most up to date subscriber to all sets ($1) on unresponsive node $2},
126 'autofailover_promote_skip' => q{No failover required for unresponsive node $1 as it is neither the origin or an active forwarder of any sets},
127 'autofailover_promote_fail' => q{Could not find suitable backup node for promotion},
128 'autofailover_node_detail' => q{Node $1 is $2 subscribed to ($3) node $4 and provides sets $5 at $6 seconds lag (on event $7)},
129 'autofailover_promote_best' => q{Best node for promotion is node $1 lag = $2 seconds (event $3)},
130 'autofailover_promote_unsuitable' => q{Node $1 is unsuitable for promotion},
131 'autofailover_unresponsive' => q{Detected unresponsive origin node: $1},
132 'autofailover_unresponsive_prov' => q{Detected unresponsive provider node: $1},
133 'autofailover_unresponsive_subonly'=> q{Detected unresponsive subscriber only node: $1},
134 'autofailover_recovery_subonly' => q{Detected recovery of previously unresponsive subscriber only node: $1},
135 'autofailover_pspec_check_fail' => q{Failed to connect to node $1: $2},
136 'autofailover_pspec_check' => q{Getting objective judgement from other nodes, apparent unresponsive nodes are : $1 (Failed nodes = $2 of $3)},
137 'autofailover_pspec_check_sleep' => q{Sleeping for $1 ms},
138 'autofailover_pspec_check_data' => q{$1: Node $2 says lag from node $3 -> $4 is $5 seconds},
139 'autofailover_pspec_check_true' => q{All detected failed nodes confirmed as lagging by other nodes},
140 'autofailover_pspec_check_false' => q{Not all nodes confirmed as lagging},
141 'autofailover_pspec_check_unknown' => q{Unable to confirm lag status of all nodes},
142 'autofailover_split_check' => q{Surviving nodes ($1 of $2) are the majority},
143 'autofailover_split_check_fail' => q{Surviving nodes ($1) are not the majority},
144 'interactive_head_id' => q{ID},
145 'interactive_head_name' => q{Name},
146 'interactive_head_status' => q{Status},
147 'interactive_head_providers' => q{Provider IDs},
148 'interactive_head_config' => q{Configuration},
149 'interactive_detail_1' => q{Origin for sets: },
150 'interactive_detail_2' => q{Providing sets: },
151 'interactive_detail_3' => q{Subscriptions: },
152 'interactive_choose_node' => q{Please choose the node to move all sets $1:},
153 'interactive_confirm' => q{You chose to move sets $1 node $2 ($3). Is this correct [y/n]? },
154 'interactive_action' => q{Best course of action is most likely to do a "$1". Do you wish to continue [y/n]?},
155 'interactive_surrender' => q{Uable to determine best course of action},
156 'interactive_write_script' => q{Writing a script to $1 node $2 to $3},
157 'interactive_check_nodes' => q{Checking availability of database nodes...},
158 'interactive_continue' => q{Do you wish to continue [y/n]?},
159 'interactive_drop_nodes' => q{Do you want to also drop the failed nodes from the slony configuration [y/n]?},
160 'interactive_preserve' => q{Preserve subscription paths to follow the origin node (choose no if unsure) [y/n]?},
161 'interactive_aliases' => q{Generate aliases based on sl_node/set comments in parentheses (choose no if unsure) [y/n]?},
162 'interactive_summary' => q{Summary of nodes to be passed to failover:},
163 'interactive_node_info' => q{Node : $1 ($2) $3 (conninfo $4)},
164 'interactive_run_script' => q{Would you like to run this script now [y/n]?},
165 'interactive_running' => q{Running the script now. This may take some time; please be patient!},
166 'interactive_reason' => q{Please enter a brief reson for taking this action: },
167 'interactive_failover_detail_1' => q{Before you go any further please consider the impact of a full failover:},
168 'interactive_failover_detail_2' => q{The node you are failing over from will cease to participate in the cluster permanently until it is rebuild & subscribed},
169 'interactive_failover_detail_3' => q{If the outage is temporary (i.e. network/power/easily replaceable hardware related) consider waiting it out},
170 'interactive_failover_detail_4' => q{This type of failover is likely to be more a business decision than a technical one},
171 'info_all_nodes_available' => q{INFO: All nodes are available},
172 'info_req_nodes_available' => q{INFO: $1 of $2 nodes are available. No unavailable nodes are subscribed to the old origin},
173 'wrn_node_unavailable' => q{WARNING: Node $1 unavailable},
174 'wrn_req_unavailable' => q{WARNING: Old origin node ($1) is available, however $2 subscribers are unavailable},
175 'wrn_not_tested' => q{WARNING: Script not tested with Slony-I v$1},
176 'wrn_failover_issues' => q{WARNING: Slony-I v$1 may struggle to failover correctly with multiple failed nodes (affects v2.0-2.1)},
177 'note_autofail_fwd_only' => q{NOTICE: Slony versions prior to 2.2 cannot initiate failover from only failed forwarding providers},
178 'note_fail_sub_only' => q{NOTICE: Slony versions prior to 2.2 cannot failover subscriber only nodes, reverting to failover_offline_subscriber_only = false},
179 'note_multiple_try' => q{NOTICE: Cannot lock multiple sets within try blocks in version $1 dropping back to single sets},
180 'note_reshape_cluster' => q{NOTICE: Either drop the failed subscribers or bring them back up, then retry to MOVE SET},
181 'dbg_generic' => q{DEBUG: $1},
182 'dbg_cluster' => q{DEBUG: NodeID $1/ProvIDs $2/Conninfo $3/OrigSets $4/NodeName $5/ProvTree $6/Active $7/FwdSets $8/ActSubSets $9},
183 'dbg_resubscribe' => q{DEBUG: Checking possibility to resubscribe set $1 ($2) to node $3 ($4) which pulls $5 ($6) from $7 ($8)},
184 'dbg_failover_method' => q{DEBUG: Failover method is $1},
185 'dbg_cluster_load' => q{DEBUG: Loading cluster configuration from $1},
186 'dbg_cluster_good' => q{DEBUG: Cluster state good},
187 'dbg_autofailover_check' => q{DEBUG: Checking node $1 ($2) role is $3 (conninfo: $4)},
188 'dbg_autofailover_active_check' => q{DEBUG: Initiate active check of $1 node $2},
189 'dbg_slonik_script' => q{DEBUG: Running slonik script $1},
190 'err_generic' => q{ERROR: $1},
191 'err_no_database' => q{ERROR: Please specify a database name},
192 'err_no_cluster' => q{ERROR: Please specify a slony cluster name},
193 'err_no_host' => q{ERROR: Please specify a host},
194 'err_no_config' => q{ERROR: No valid config found},
195 'err_fail_config' => q{ERROR: Failed to load configuration},
196 'err_write_fail' => q{ERROR: Could not write to $1 "$2"},
197 'err_read_fail' => q{ERROR: Could not read from $1 "$2"},
198 'err_unlink_fail' => q{ERROR: Could not delete $1 "$2"},
199 'err_mkdir_fail' => q{ERROR: Unable to create $1 directory "$2"},
200 'err_execute_fail' => q{ERROR: Could not execute $1 "$2"},
201 'err_inactive' => q{ERROR: Node $1 is not active (state = $2)},
202 'err_cluster_empty' => q{ERROR: Loaded cluster contains no nodes},
203 'err_cluster_offline' => q{ERROR: Loaded cluster contains no reachable nodes},
204 'err_cluster_lone' => q{ERROR: Loaded cluster contains only 1 node},
205 'err_not_origin' => q{ERROR: Node $1 is not the origin of any sets},
206 'err_not_provider' => q{ERROR: Node $1 is not a provider of any sets},
207 'err_not_provider_sets' => q{ERROR: Node $1 does not provide the sets required: need ($2) but provides ($3)},
208 'err_no_configuration' => q{ERROR: Could not read configuration for node $1},
209 'err_must_enter_node_id' => q{ERROR: You must enter a node id},
210 'err_not_a_node_id' => q{ERROR: I have no knowledge of a node $1},
211 'err_same_node' => q{ERROR: Cant move from and to the same node},
212 'err_node_offline' => q{ERROR: $1 node ($2) is not available},
213 'err_incomplete_preamble' => q{ERROR: Incomplete preamble},
214 'err_running_slonik' => q{ERROR: Could not run slonik: $1},
215 'err_pgsql_connect' => q{ERROR: Cannot connect to postgres server},
216 'slonik_output' => q{SLONIK: $1},
217 'exit_noaction' => q{Exiting, no action has been taken},
218 'exit' => q{Exited by $1}
221 'usage' => q{-h <host> -p <port> -db <database> -cl <cluster name> -u <username> -P <password> -f <config file> (Option mot de passe pas recommandé; utiliser pgpass place)},
222 'title' => q{Slony-I failover (basculement) version de script $1},
223 'cluster_fixed' => q{Abandon de l'action de basculement: tous les noeuds d'origine / de fournisseurs maintenant sensible},
224 'cluster_failed' => q{Trouvé $1 échoué noeuds, couchage pour $2 ms avant réessayer $3 de $4},
225 'load_cluster' => q{Obtenir une liste de noeuds de base de donnees...},
226 'load_cluster_fail' => q{Impossible de lire la configuration du cluster $1},
227 'load_cluster_success' => q{Chargé Slony-I v$1 groupe "$2" avec $3 noeuds lire à partir du noeud à $4:$5/$6},
228 'lag_detail' => q{Current informations noeud de décalage à partir du noeud de configuration:},
229 'script_settings' => q{Utilisation de $1 lots de système de verrouillage, $2 FAILOVER et $3},
230 'generated_script' => q{Script généré "$1"},
231 'autofailover_init' => q{Entrer dans le mode de autofailover},
232 'autofailover_init_cnf' => q{Configuration Slony sera lu à partir de $1 noeud},
233 'autofailover_init_pol' => q{Vérifier toutes les $1ms},
234 'autofailover_init_ret' => q{Noeuds défaillants seront rejugés $1 fois avec $2 ms sommeil},
235 'autofailover_init_set' => q{Fournisseurs d'expédition échoué $1 être échoué sur},
236 'autofailover_load_cluster' => q{$1 Slony-I v$2 grappe "$3" avec $4 noeuds lire à noeud $5},
237 'autofailover_proceed' => q{De procéder à failover:},
238 'autofailover_detail' => q{Noeud défaillant: $1, noeud de sauvegarde: $2},
239 'autofailover_halt' => q{Noeuds Impossible d'effectuer une failover pour $1 échoué},
240 'autofailover_check_sub' => q{Vérification noeud abonné $1},
241 'autofailover_check_sub_fail' => q{Impossible de vérifier noeud abonné $1},
242 'autofailover_promote_find' => q{Trouver plus à jour abonné à tous les jeux ($1) sur le noeud ne répond pas $2},
243 'autofailover_promote_found' => q{Utilisation précédemment trouvé plus à jour abonné à tous les jeux ($1) sur le noeud ne répond pas $2},
244 'autofailover_promote_skip' => q{Pas de failover requis pour le noeud ne répond pas $1 car il n'est ni l'origine ou un transitaire active de tous les jeux},
245 'autofailover_promote_fail' => q{Impossible de trouver le noeud de sauvegarde approprié pour la promotion},
246 'autofailover_node_detail' => q{Noeud $1 est souscrit à $2 ($3) noeud $4 et fournit des ensembles de $5 à retard $6 secondes (en cas d'événement $7)},
247 'autofailover_promote_best' => q{Meilleur noeud pour la promotion est noeud $1 décalage = $2 secondes (événement $3)},
248 'autofailover_promote_unsuitable' => q{Noeud $1 est inadapté pour la promotion},
249 'autofailover_unresponsive' => q{Noeud d'origine ne répond pas détecté: $1},
250 'autofailover_unresponsive_prov' => q{Noeud fournisseur ne répond pas détecté: $1},
251 'autofailover_unresponsive_subonly'=> q{Abonné ne répond pas détecté seulement de noeud: $1},
252 'autofailover_recovery_subonly' => q{Recouvrement détecté de l'abonné ne répond pas seulement auparavant de noeud: $1},
253 'autofailover_pspec_check_fail' => q{Impossible de se connecter au noeud $1: $2},
254 'autofailover_pspec_check' => q{Obtenir un jugement objectif à partir d'autres noeuds, les noeuds qui ne répondent pas apparentes sont : $1 (Noeuds défaillants = $2 de $3)},
255 'autofailover_pspec_check_sleep' => q{Dormir pour $1 ms},
256 'autofailover_pspec_check_data' => q{$1: Noeud $2 dit décalage de $3 -> $4 noeud est $5 secondes},
257 'autofailover_pspec_check_true' => q{Tous les noeuds détectés pas confirmés comme à la traîne par d'autres noeuds},
258 'autofailover_pspec_check_false' => q{Pas tous les noeuds confirmé retard},
259 'autofailover_pspec_check_unknown' => q{Impossible de confirmer le statut de latence de tous les noeuds},
260 'autofailover_split_check' => q{Autres noeuds ($1 sur $2) sont la majorité},
261 'autofailover_split_check_fail' => q{Autres noeuds ($1) ne sont pas la majorité},
262 'interactive_head_name' => q{Nom},
263 'interactive_head_status' => q{Statut},
264 'interactive_head_providers' => q{Fournisseur IDs},
265 'interactive_detail_1' => q{Origine pour les jeux: },
266 'interactive_detail_2' => q{Fournir des ensembles: },
267 'interactive_detail_3' => q{Abonnements: },
268 'interactive_choose_node' => q{S'il vous plaît choisissez le noeud à déplacer tous les ensembles $1:},
269 'interactive_confirm' => q{Vous avez choisi de passer ensembles $1 noeud $2 ($3). Est-ce correct [o/n]? },
270 'interactive_drop_nodes' => q{Voulez-vous laisser tomber aussi les noeuds défaillants de la configuration de slony [o/n]?},
271 'interactive_action' => q{Meilleur plan d'action est le plus susceptible de faire une $1. Voulez-vous continuer [o/n]?},
272 'interactive_surrender' => q{Uable pour déterminer le meilleur plan d'action},
273 'interactive_write_script' => q{Rédaction d'un script à $1 $2 noeud à $3},
274 'interactive_check_nodes' => q{Vérification de la disponibilité des noeuds de base de donnees...},
275 'interactive_continue' => q{Voulez-vous continuer [o/n]?},
276 'interactive_preserve' => q{Préserver les chemins de souscription à suivre le noeud d'origine (ne pas choisir en cas de doute) [o/n]?},
277 'interactive_aliases' => q{Générer des alias sur la base de sl_node / set commentaires entre parenthèses (ne pas choisir en cas de doute) [o/n]?},
278 'interactive_summary' => q{Résumé des noeuds à passer à failover:},
279 'interactive_node_info' => q{Noeud : $1 ($2) $3 (conninfo $4)},
280 'interactive_run_script' => q{Voulez-vous exécuter ce script maintenant [o/n]?},
281 'interactive_running' => q{L'exécution du script maintenant. Cela peut prendre un certain temps; s'il vous plaît être patient!},
282 'interactive_reason' => q{S'il vous plaît entrer une brève reson pour cette action: },
283 'interactive_failover_detail_1' => q{Avant d'aller plus loin s'il vous plaît envisager l'impact d'un failover (basculement) complet:},
284 'interactive_failover_detail_2' => q{Le noeud vous ne parviennent pas au-dessus de cesse de participer au groupe de façon permanente jusqu'à ce qu'il soit à reconstruire et souscrit},
285 'interactive_failover_detail_3' => q{Si la panne est temporaire (c.-à -réseau / alimentation / facilement remplaçable matériel connexe) envisager d'attendre dehors},
286 'interactive_failover_detail_4' => q{Ce type de failover est susceptible d'être plus une décision d'affaires que technique},
287 'info_all_nodes_available' => q{INFO: Tous les noeuds sont disponibles},
288 'info_req_nodes_available' => q{INFO: $1 of $2 noeuds sont disponibles. Pas de noeuds indisponibles sont souscrites à l'ancienne origine},
289 'wrn_node_unavailable' => q{ATTENTION: Noeud $1 disponible},
290 'wrn_req_unavailable' => q{ATTENTION: Noeud Old origine ($1) est disponible, mais $2 abonnés ne sont pas disponibles},
291 'wrn_not_tested' => q{ATTENTION: Script pas testé avec Slony-I v$1},
292 'wrn_failover_issues' => q{ATTENTION: Slony-I v$1 peut lutter pour basculer correctement avec plusieurs nœuds défaillants (affecte v2.0-2.1)},
293 'note_autofail_fwd_only' => q{AVIS: Versions antérieures à la 2.2 Slony ne peuvent pas initier le basculement de seulement échoué transmettre fournisseurs},
294 'note_fail_sub_only' => q{AVIS: Versions antérieures à la 2.2 Slony ne peuvent pas basculer abonnes seuls les noeuds, revenant à failover_offile_subscriber_only = false},
295 'note_multiple_try' => q{AVIS: Vous ne pouvez pas verrouiller plusieurs ensembles dans des blocs try dans la version $1 de retomber à des jeux simples},
296 'note_reshape_cluster' => q{AVIS: Vous devez supprimer les abonnés défaillants ou les ramener, puis réessayez à MOVE SET},
297 'err_generic' => q{ERREUR: $1},
298 'err_no_database' => q{ERREUR: S'il vous plaît spécifier un base de donnees nom},
299 'err_no_cluster' => q{ERREUR: S'il vous plaît indiquez un nom de cluster slony},
300 'err_no_host' => q{ERREUR: S'il vous plaît spécifier un hôte},
301 'err_no_config' => q{ERREUR: Aucune configuration valide n'a été trouvée},
302 'err_fail_config' => q{ERREUR: Impossible de charger la configuration},
303 'err_write_fail' => q{ERREUR: Impossible d'écrire dans $1 "$2"},
304 'err_read_fail' => q{ERREUR: Impossible de lire $1 "$2"},
305 'err_unlink_fail' => q{ERREUR: Impossible de supprimer $1 "$2"},
306 'err_mkdir_fail' => q{ERREUR: Impossible de créer $1 répertoire "$2"},
307 'err_execute_fail' => q{ERREUR: Impossible d'exécuter $1 "$2"},
308 'err_inactive' => q{ERREUR: Noeud $1 n'est pas active (état = $2)},
309 'err_cluster_empty' => q{ERREUR: Groupe chargé contient pas de noeuds},
310 'err_cluster_offline' => q{ERREUR: Groupe chargé contient pas de noeuds accessibles},
311 'err_cluster_lone' => q{ERRRUE: Groupe chargé ne contient que 1 noeud},
312 'err_not_origin' => q{ERREUR: Noeud $1 n'est pas à l'origine de tous les jeux},
313 'err_not_provider' => q{ERREUR: Noeud $1 n'est pas un fournisseur de tous les jeux},
314 'err_not_provider_sets' => q{ERREUR: Noeud $1 ne fournit pas les ensembles nécessaires: le besoin ($2), mais fournit ($3)},
315 'err_no_configuration' => q{ERREUR: Impossible de lire la configuration pour le noeud $1},
316 'err_must_enter_node_id' => q{ERREUR: Vous devez entrer un id de noeud},
317 'err_not_a_node_id' => q{ERREUR: Je n'ai pas connaissance d'un $1 de noeud},
318 'err_same_node' => q{ERREUR: Cant déplacer depuis et vers le même noeud},
319 'err_node_offline' => q{ERREUR: $1 noeud ($2) n'est pas disponible},
320 'err_incomplete_preamble' => q{ERREUR: Préambule incomplète},
321 'err_running_slonik' => q{ERREUR: Ne pouvait pas courir slonik: $1},
322 'err_pgsql_connect' => q{ERREUR: Impossible de se connecter au serveur postgres},
323 'slonik_output' => q{SLONIK: $1},
324 'exit_noaction' => q{Quitter, aucune action n'a été prise},
325 'exit' => q{Quitter par $1}
330 # Setup date variables
331 my ($g_year, $g_month, $g_day, $g_hour, $g_min, $g_sec) = (localtime(time))[5,4,3,2,1,0];
332 my $g_date = sprintf ("%02d:%02d:%02d on %02d/%02d/%04d", $g_hour, $g_min, $g_sec, $g_day, $g_month+1, $g_year+1900);
334 # Handle command line options
335 Getopt::Long::Configure('no_ignore_case');
337 die lookupMsg('usage') unless GetOptions(\%opt, 'host|H=s', 'port|p=i', 'dbname|db=s', 'clname|cl=s', 'dbuser|u=s', 'dbpass|P=s', 'cfgfile|f=s', 'infoprint|I', ) and keys %opt and ! @ARGV;
340 if (defined($opt{cfgfile})) {
341 unless (getConfig($opt{cfgfile})) {
342 println(lookupMsg('err_no_config'));
347 if (defined($opt{dbname})) {
348 $g_dbname = $opt{dbname};
350 if (defined($opt{clname})) {
351 $g_clname = $opt{clname};
353 if (defined($opt{host})) {
354 $g_dbhost = $opt{host};
356 if (defined($opt{port})) {
357 $g_dbport = $opt{port};
359 if (defined($opt{dbuser})) {
360 $g_dbuser = $opt{dbuser};
362 if (defined($opt{dbpass})) {
363 $g_dbpass = $opt{dbpass};
367 # Display message and die if any of the required configuration variables are missing
368 if (!defined($g_dbname)) {
369 println(lookupMsg('err_no_database'));
370 die lookupMsg('usage');
372 if (!defined($g_clname)) {
373 println(lookupMsg('err_no_cluster'));
374 die lookupMsg('usage');
376 if (!defined($g_dbhost)) {
377 println(lookupMsg('err_no_host'));
378 die lookupMsg('usage');
382 # Build conninfo from supplied datbase name/host/port
383 $g_dbconninfo = "dbname=$g_dbname;host=$g_dbhost;port=$g_dbport";
385 if (!defined($opt{infoprint})) {
386 # Check prefix directory and create if not present
387 unless(-e $g_prefix or mkdir $g_prefix) {
388 println(lookupMsg('err_mkdir_fail', 'prefix', $g_prefix));
392 if ($g_separate_working) {
393 if ($g_prefix !~ m/\/$/) {
397 # Get a uuid for working directory
398 $g_prefix .= getUUID($g_date);
400 # Create a working directory and setup log file
401 unless(-e $g_prefix or mkdir $g_prefix) {
402 println(lookupMsg('err_mkdir_fail', 'work', $g_prefix));
407 # Set postgres path if provided
408 if (defined($g_slonikpath) && ($g_slonikpath ne "")) {
409 $ENV{PATH} .= ":$g_slonikpath";
412 # Check if autofailover is enabled, if so check configuration and enter autofailover mode
413 if (($g_autofailover) && !defined($opt{infoprint})) {
415 # Write out a PID file
416 if (writePID($g_prefix, $g_logfile, $g_log_prefix, $g_pidfile)) {
417 $g_pid_written = true;
420 cleanExit(1, "system");
423 # Go into endless loop for autofailover
424 autoFailover($g_dbconninfo, $g_clname, $g_dbuser, $g_dbpass, $g_prefix, $g_logfile, $g_log_prefix);
427 # Read slony configuration and output some basic information
430 println(lookupMsg('load_cluster', $g_prefix));
431 ($g_node_count, $g_version) = loadCluster($g_dbconninfo, $g_clname, $g_dbuser, $g_dbpass, $g_prefix, $g_logfile, $g_log_prefix);
434 println(lookupMsg('load_cluster_fail', 'from supplied configuration'));
435 cleanExit(2, "system");
438 if (defined($opt{infoprint})) {
439 println(lookupMsg('load_cluster_success', $g_version, $g_clname, $g_node_count, $g_dbhost, $g_dbport, $g_dbname) . ":");
440 chooseNode("info", undef, undef, undef, 0);
444 printlog($g_prefix,$g_logfile,$g_log_prefix,"*"x68 . "\n* ");
445 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('title', $g_script_version));
446 printlogln($g_prefix,$g_logfile,$g_log_prefix,"*"x68);
449 if ($g_node_count <= 0) {
450 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('err_cluster_empty'));
451 cleanExit(3, "system");
454 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('load_cluster_success', $g_version, $g_clname, $g_node_count, $g_dbhost, $g_dbport, $g_dbname));
455 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('script_settings', $g_lockset_method, $g_failover_method, uc($g_resubscribe_method)));
458 # Output lag information between each node and node configuration was read from
459 if (loadLag($g_dbconninfo, $g_clname, $g_dbuser, $g_dbpass, $g_prefix, $g_logfile, $g_log_prefix) > 0) {
460 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('lag_detail'));
462 printlogln($g_prefix,$g_logfile,$g_log_prefix,"\t$_");
464 printlog($g_prefix,$g_logfile,$g_log_prefix,"\n");
467 # Prompt user to choose nodes to move sets from / to
468 $g_node_from = chooseNode("from", $g_prefix, $g_logfile, $g_log_prefix, 0);
469 if ($g_node_from == 0) {
470 cleanExit(4, "user");
472 elsif ($g_node_from == -1) {
473 cleanExit(5, "system");
476 $g_node_to = chooseNode("to", $g_prefix, $g_logfile, $g_log_prefix, $g_node_from);
477 if ($g_node_to == 0) {
478 cleanExit(6, "user");
480 elsif ($g_node_to == -1) {
481 cleanExit(7, "system");
483 elsif ($g_node_from == $g_node_to) {
484 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('err_same_node'));
485 cleanExit(8, "system");
488 # Check nodes are available and decide on action to take
489 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_check_nodes'));
490 ($g_available_node_count, $g_critical_node_count) = checkNodes($g_clname, $g_dbuser, $g_dbpass, $g_node_from, $g_node_to, $g_prefix, $g_logfile, $g_log_prefix);
492 if ($g_available_node_count <= 0) {
493 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('err_cluster_offline'));
494 cleanExit(9, "system");
496 elsif ($g_critical_node_count == -1) {
497 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('err_node_offline', 'Target new origin', $g_node_to));
498 cleanExit(10, "system");
500 elsif ($g_critical_node_count == -2) {
501 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('err_node_offline', 'Old origin', $g_node_from));
502 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_action', 'FAILOVER'));
505 elsif ($g_critical_node_count == 0) {
506 if ($g_node_count == $g_available_node_count) {
507 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('info_all_nodes_available'));
510 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('info_req_nodes_available', $g_available_node_count, $g_node_count));
512 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_action', 'MOVE SET'));
514 elsif ($g_critical_node_count > 0) {
515 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('wrn_req_unavailable', $g_node_from, $g_critical_node_count));
516 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('note_reshape_cluster'));
517 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('exit_noaction'));
518 cleanExit(11, "user");
521 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_surrender'));
522 cleanExit(12, "system");
526 if ($g_input !~ /^[Y|O]$/i) {
527 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('exit_noaction'));
528 cleanExit(13, "user");
531 if (!$g_use_comment_aliases) {
532 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_aliases'));
535 if ($g_input =~ /^[Y|O]$/i) {
536 $g_use_comment_aliases = true;
541 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_summary'));
543 foreach (@g_failed) {
544 printlogln($g_prefix,$g_logfile,$g_log_prefix,"\t" . lookupMsg('interactive_node_info',$_->[0],($_->[4] // "unnamed"),(defined($_->[9]) ? "providing sets $_->[9]" : "sole subscriber"), $_->[2]));
547 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_failover_detail_1'));
548 printlogln($g_prefix,$g_logfile,$g_log_prefix,"\t" . lookupMsg('interactive_failover_detail_2'));
549 printlogln($g_prefix,$g_logfile,$g_log_prefix,"\t" . lookupMsg('interactive_failover_detail_3'));
550 printlogln($g_prefix,$g_logfile,$g_log_prefix,"\t" . lookupMsg('interactive_failover_detail_4'));
552 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_drop_nodes'));
554 if ($g_input ~~ /^[Y|O]$/i) {
555 $g_drop_failed = true;
558 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_reason'));
560 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_continue'));
563 if ($g_input !~ /^[Y|O]$/i) {
564 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('exit_noaction'));
565 cleanExit(14, "user");
568 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_write_script', 'failover from', $g_node_from, $g_node_to));
569 $g_script = writeFailover($g_prefix, $g_dbconninfo, $g_clname, $g_dbuser, $g_dbpass, $g_node_from, $g_node_to, $g_subs_follow_origin, $g_use_comment_aliases, $g_logfile, $g_log_prefix);
572 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_preserve'));
575 if ($g_input =~ /^[Y|O]$/i) {
576 $g_subs_follow_origin = true;
579 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_reason'));
582 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_write_script', 'move all sets provided by', $g_node_from, $g_node_to));
583 $g_script = writeMoveSet($g_prefix, $g_dbconninfo, $g_clname, $g_dbuser, $g_dbpass, $g_node_from, $g_node_to, $g_subs_follow_origin, $g_use_comment_aliases, $g_logfile, $g_log_prefix);
586 # Complete and run script if required
588 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('generated_script', $g_script));
589 printlog($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_run_script', $g_script));
592 if ($g_input =~ /^[Y|O]$/i) {
593 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('interactive_running'));
594 unless (runSlonik($g_script, $g_prefix, $g_logfile, $g_log_prefix)) {
595 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('err_execute_fail', 'slonik script', $g_script));
599 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('exit_noaction'));
603 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('err_read_fail', 'slonik script', $g_script));
604 cleanExit(15, "system");
607 cleanExit(0, "script completion");
609 ###########################################################################################################################################
611 # Display exit message, insert log file into database if requested, delete any pid files and exit with the requested code
613 my $exit_code = shift;
616 printlogln($g_prefix,$g_logfile,$g_log_prefix,lookupMsg('exit', $type));
620 logDB("dbname=$g_logdb_name;host=$g_logdb_host;port=$g_logdb_port", $g_logdb_user, $g_logdb_pass, $exit_code, $g_reason, $g_prefix, $g_logfile, $g_log_prefix, $g_clname, $g_script);
624 if ($g_pid_written) {
625 removePID($g_prefix, $g_logfile, $g_log_prefix, $g_pidfile);
631 # Exit on caught signal
633 cleanExit(100,'signal');
636 # Check we can reach each node in the cluster and that it contains the Slony schema
645 my $log_prefix = shift;
651 my $result_count = 0;
652 my $critical_count = 0;
658 undef @g_unresponsive;
661 foreach (@g_cluster) {
662 if ($_->[0] == $from) {
663 @origsets = split(',', $_->[3]);
668 foreach (@g_cluster) {
670 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_cluster', $_->[0],($_->[1] // "<NONE>"),$_->[2],($_->[3] // "<NONE>"),$_->[4],($_->[5] // "<NONE>") . "(" . ($_->[8] // "<NONE>") . ")",$_->[6],($_->[7] // "<NONE>"),($_->[9] // "<NONE>") . " (" . ($_->[10] // "<NONE>") . ")"));
673 $dsn = "DBI:Pg:$_->[2];";
675 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
676 $query = "SELECT count(*) FROM pg_namespace WHERE nspname = ?";
677 $sth = $dbh->prepare($query);
678 $sth->bind_param(1, "_" . $clname);
681 $result_count = $result_count+$sth->rows;
688 # Critical count will be -1 if the new origin is down, -2 if the old origin is down or positive if subscribers to sets on old origin are down.
689 printlogln($prefix,$logfile,$log_prefix,lookupMsg('wrn_node_unavailable', $_->[0]));
691 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
693 if ($_->[0] == $to) {
694 $critical_count = -1;
696 elsif ($_->[0] == $from) {
697 $critical_count = -2;
700 foreach my $subprov (split(';', $_->[5])) {
701 my ($node, $setlist) = (split('->', $subprov)) ;
703 $setlist =~ s/(\)|\(|s)//g;
704 @subsets = (split(',', $setlist));
706 if (($critical_count >= 0) && (checkSubscribesAnySets(\@origsets, \@subsets))) {
711 # Only push nodes with active subscribers to sets into failed list unless explicitly told to
712 if (($g_fail_subonly) || (defined($_->[9]))) {
713 push(@g_failed, \@$_);
714 $g_backups{$_->[0]} = $to;
716 push(@g_unresponsive, \@$_);
720 return ($result_count, $critical_count);
723 # Load information on all nodes in the Slony cluster into global @g_cluster:
724 # 0) no_id = Node id of this node
725 # 1) no_provs = Comma separated list of all provider node ids
726 # 2) no_conninfo = Conninfo as recorded in sl_path
727 # 3) origin_sets = Comma separated list of set ids originating on this node
728 # 4) no_name = Node name; this is extracted from text between parentheses in sl_node.no_comment
729 # 5) no_sub_tree = Text representation of subscriptions in the form n<provider node id>->(s<set id>, ..)
730 # 6) no_status = Text representing the state of the node; either ACTIVE,INACTIVE or FAILED
731 # 7) sub_sets = Comma separated list of all set ids this node is subscribed to
732 # 8) no_sub_tree_name = As per no_sub_tree but holds textual names extracted from sl_node.no_comment
733 # 9) prov_sets_active = Comma separated list of all set ids this node is actively forwarding
734 # 10) prov_sets = Comma separated list of all set ids this node is subscribed to and able to forward
736 my $dbconninfo = shift;
742 my $log_prefix = shift;
753 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_cluster_load', $dbconninfo));
756 $dsn = "DBI:Pg:$dbconninfo;";
758 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
759 $qw_clname = $dbh->quote_identifier("_" . $clname);
761 $query = "SELECT $qw_clname.getModuleVersion()";
762 $sth = $dbh->prepare($query);
764 ($version) = $sth->fetchrow;
767 $query = "WITH z AS (
768 SELECT a.no_id, b.sub_provider AS no_prov,
769 COALESCE(c.pa_conninfo,(SELECT pa_conninfo FROM $qw_clname.sl_path WHERE pa_server = $qw_clname.getlocalnodeid(?) LIMIT 1)) AS no_conninfo,
770 array_to_string(array(SELECT set_id FROM $qw_clname.sl_set WHERE set_origin = a.no_id ORDER BY set_id),',') AS origin_sets,
771 string_agg(CASE WHEN b.sub_receiver = a.no_id AND b.sub_forward AND b.sub_active THEN b.sub_set::text END, ',' ORDER BY b.sub_set) AS sub_sets,
772 coalesce(trim(regexp_replace(substring(a.no_comment from E'\\\\((.+)\\\\)'), '[^0-9A-Za-z]','_','g')), 'node' || a.no_id) AS no_name,
773 'n' || b.sub_provider || '->(' || string_agg(CASE WHEN b.sub_receiver = a.no_id THEN 's' || b.sub_set END,',' ORDER BY b.sub_set,',') || ')' AS sub_tree,
774 coalesce(trim(regexp_replace(substring(d.no_comment from E'\\\\((.+)\\\\)'), '[^0-9A-Za-z]','_','g')), 'node' || b.sub_provider, '')
775 || '->(' || string_agg(CASE WHEN b.sub_receiver = a.no_id THEN coalesce(trim(regexp_replace(e.set_comment, '[^0-9A-Za-z]', '_', 'g')), 'set' || b.sub_set) END,',' ORDER BY b.sub_set) || ')' AS sub_tree_name,
776 CASE " . ((substr($version,0,3) >= 2.2) ? "WHEN a.no_failed THEN 'FAILED' " : "") . "WHEN a.no_active THEN 'ACTIVE' ELSE 'INACTIVE' END AS no_status,
777 array_to_string(array(SELECT DISTINCT sub_set::text FROM $qw_clname.sl_subscribe WHERE sub_provider = a.no_id AND sub_active ORDER BY sub_set),',') AS prov_sets_active,
778 string_agg(CASE WHEN b.sub_receiver = a.no_id THEN b.sub_set::text END,',' ORDER BY b.sub_set,',') AS prov_sets
779 FROM $qw_clname.sl_node a
780 LEFT OUTER JOIN $qw_clname.sl_subscribe b ON a.no_id = b.sub_receiver
781 LEFT OUTER JOIN $qw_clname.sl_path c ON c.pa_server = a.no_id AND c.pa_client = $qw_clname.getlocalnodeid(?)
782 LEFT OUTER JOIN $qw_clname.sl_node d ON b.sub_provider = d.no_id
783 LEFT OUTER JOIN $qw_clname.sl_set e ON b.sub_set = e.set_id
784 GROUP BY b.sub_provider, a.no_id, a.no_comment, c.pa_conninfo, d.no_comment, a.no_active
788 nullif(string_agg(no_prov::text, ',' ORDER BY no_prov),'') AS no_provs,
790 nullif(string_agg(origin_sets::text, ',' ORDER BY origin_sets),'') AS origin_sets,
792 nullif(string_agg(sub_tree, ';' ORDER BY sub_tree),'') AS no_sub_tree,
794 nullif(string_agg(sub_sets::text, ',' ORDER BY prov_sets),'') AS sub_sets,
795 nullif(string_agg(sub_tree_name, ';' ORDER BY sub_tree_name),'') AS no_sub_tree_name,
796 nullif(string_agg(prov_sets_active::text, ',' ORDER BY prov_sets_active),'') AS prov_sets_active,
797 nullif(string_agg(prov_sets::text, ',' ORDER BY sub_sets),'') AS prov_sets
798 FROM z GROUP BY no_id, no_conninfo, no_name, no_status";
799 $sth = $dbh->prepare($query);
801 $sth->bind_param(1, "_" . $clname);
802 $sth->bind_param(2, "_" . $clname);
806 while (my @node = $sth->fetchrow) {
807 #printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', join(' - ', @node)));
808 push(@g_cluster, \@node);
817 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
819 die lookupMsg('err_pgsql_connect');
822 #if (substr($version,0,1) < 2) {
823 # printlogln($prefix,$logfile,$log_prefix,lookupMsg('wrn_not_tested', $version));
825 if (($g_use_try_blocks) && ($g_lockset_method eq 'multiple') && (substr($version,0,3) <= 9.9)) {
826 # It's currently not possible to lock multiple sets at a time within a try block (v2.2.2), leave the logic in and set a high version number for now.
827 printlogln($prefix,$logfile,$log_prefix, lookupMsg('note_multiple_try', $version));
828 $g_lockset_method = 'single';
830 if (substr($version,0,3) >= 2.2) {
831 $g_failover_method = 'new';
832 $g_resubscribe_method = 'resubscribe';
835 unless ($g_silence_notice) {
836 if ((substr($version,0,3) >= 2.0) && (substr($version,0,3) < 2.2)) {
837 printlogln($prefix,$logfile,$log_prefix,lookupMsg('wrn_failover_issues', $version));
839 printlogln($prefix,$logfile,$log_prefix,lookupMsg('note_autofail_fwd_only'));
840 $g_silence_notice = true;
842 if ($g_fail_subonly) {
843 printlogln($prefix,$logfile,$log_prefix,lookupMsg('note_fail_sub_only'));
844 $g_fail_subonly = false;
850 return (scalar(@g_cluster), $version);
853 # Load all sets originating on a node into global @g_sets
855 my $dbconninfo = shift;
857 my $nodenumber = shift;
862 my $log_prefix = shift;
872 $dsn = "DBI:Pg:$dbconninfo;";
874 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
875 $qw_clname = $dbh->quote_identifier("_" . $clname);
876 $query = "SELECT set_id, trim(regexp_replace(set_comment,'[^0-9,A-Z,a-z]','_','g')) FROM $qw_clname.sl_set WHERE set_origin = ? ORDER BY set_id;";
878 $sth = $dbh->prepare($query);
879 $sth->bind_param(1, $nodenumber);
883 while (my @set = $sth->fetchrow) {
884 push(@g_sets, \@set);
892 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
894 die lookupMsg('err_pgsql_connect');
897 return scalar(@g_sets);
900 # Load information regarding replication lag from sl_status into @g_lags
901 # If loading from a node that is not the intended origin then this information might not be that accurate/useful
903 my $dbconninfo = shift;
909 my $log_prefix = shift;
919 $dsn = "DBI:Pg:$dbconninfo;";
921 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
922 $qw_clname = $dbh->quote_identifier("_" . $clname);
923 $query = "SELECT a.st_origin || ' (' || coalesce(trim(regexp_replace(substring(b.no_comment from E'\\\\((.+)\\\\)'), '[^0-9A-Za-z]','_', 'g')), 'node' || b.no_id) || ')<->'
924 || a.st_received || ' (' || coalesce(trim(regexp_replace(substring(c.no_comment from E'\\\\((.+)\\\\)'), '[^0-9A-Za-z]','_', 'g')), 'node' || c.no_id) || ') Events: '
925 || a.st_lag_num_events || ' Time: ' || a.st_lag_time
926 FROM $qw_clname.sl_status a
927 INNER JOIN $qw_clname.sl_node b on a.st_origin = b.no_id
928 INNER JOIN $qw_clname.sl_node c on a.st_received = c.no_id";
930 $sth = $dbh->prepare($query);
933 while (my $lag = $sth->fetchrow) {
942 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
944 die lookupMsg('err_pgsql_connect');
947 return scalar(@g_lags);
950 # Prompt user for nodes to an from in interactive mode and do some checking
955 my $log_prefix = shift;
956 my $last_choice = shift;
965 $line = sprintf "%-4s %-14s %-10s %-24s %-s\n", lookupMsg('interactive_head_id'), lookupMsg('interactive_head_name'), lookupMsg('interactive_head_status'), lookupMsg('interactive_head_providers'), lookupMsg('interactive_head_config');
966 printlog($prefix,$logfile,$log_prefix,"$line");
967 $line = sprintf "%-4s %-14s %-10s %-24s %-s\n", "="x(length(lookupMsg('interactive_head_id'))), "="x(length(lookupMsg('interactive_head_name'))), "="x(length(lookupMsg('interactive_head_status'))), "="x(length(lookupMsg('interactive_head_providers'))), "="x(length(lookupMsg('interactive_head_config')));
968 printlog($prefix,$logfile,$log_prefix,"$line");
970 foreach (@g_cluster) {
971 $line = sprintf "%-4s %-14s %-10s %-24s %-s\n", $_->[0], $_->[4], $_->[6], ($_->[1] // "<NONE>"), (lookupMsg('interactive_detail_1') . ($_->[3] // "<NONE>"));
972 printlog($prefix,$logfile,$log_prefix,"$line");
973 $line = sprintf "%-55s %-s\n", " ", (lookupMsg('interactive_detail_2') . ($_->[7] // "<NONE>"));
974 printlog($prefix,$logfile,$log_prefix,"$line");
975 $line = sprintf "%-55s %-s\n", " ", (lookupMsg('interactive_detail_3') . ($_->[5] // "<NONE>"));
976 printlogln($prefix,$logfile,$log_prefix,"$line");
977 $options{$_->[0]} = {name => $_->[4], sets => ($_->[3] // ""), status => $_->[6], provider => $_->[7]};
979 if ($type !~ m/info/i) {
980 printlog($prefix,$logfile,$log_prefix,lookupMsg('interactive_choose_node', $type));
984 if(exists($options{$choice})) {
985 if ($options{$choice}->{status} ne "ACTIVE") {
986 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_inactive', $choice, lc($options{$choice}->{status})));
989 elsif (($type =~ m/from/i) && (length(trim($options{$choice}->{sets})) <= 0)) {
990 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_not_origin', $choice));
993 elsif ($type =~ m/to/i) {
994 if (length(trim($options{$choice}->{provider})) <= 0) {
995 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_not_provider', $choice));
999 foreach my $old_origin (@g_cluster) {
1000 if ($old_origin->[0] == $last_choice) {
1001 @sets_from = split(',', $old_origin->[3]);
1002 @sets_to = split(',', $options{$choice}->{provider});
1003 if (checkProvidesAllSets(\@sets_from, \@sets_to)) {
1007 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_not_provider_sets',$choice,$old_origin->[3],$options{$choice}->{providers}));
1014 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_no_configuration', $last_choice));
1020 printlog($prefix,$logfile,$log_prefix,lookupMsg('interactive_confirm',$type,$choice,$options{$choice}->{name}));
1023 if ($ok !~ /^[Y|O]$/i) {
1024 printlogln($prefix,$logfile,$log_prefix,lookupMsg('exit_noaction'));
1029 elsif (!length($choice)) {
1030 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_must_enter_node_id'));
1034 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_not_a_node_id', $choice));
1042 # Write a slonik preamble section using information pulled into @g_cluster and @g_sets by loadCluster() and loadSets() functions
1044 my $filename = shift;
1045 my $dbconninfo = shift;
1050 my $aliases = shift;
1052 my $logfile = shift;
1053 my $log_prefix = shift;
1054 my $comment_all_failed = shift;
1057 my $success = false;
1059 my ($year, $month, $day, $hour, $min, $sec) = (localtime(time))[5,4,3,2,1,0];
1060 my $date = sprintf ("%02d:%02d:%02d on %02d/%02d/%04d", $hour, $min, $sec, $day, $month+1, $year+1900);
1062 if (open(SLONFILE, ">", $filename)) {
1063 print SLONFILE ("# Script autogenerated on $date\n\n");
1064 print SLONFILE ("######\n# Preamble (cluster structure)\n######\n\n# Cluster name\n");
1066 print SLONFILE ("DEFINE slony_cluster_name $clname;\n");
1067 print SLONFILE ("CLUSTER NAME = \@slony_cluster_name;\n\n");
1070 print SLONFILE ("CLUSTER NAME = $clname;\n\n");
1072 foreach (@g_cluster) {
1074 if (($comment_all_failed) && (exists $g_backups{$_->[0]})) {
1075 $line_prefix = "# (Node $_->[0] unavailable) ";
1077 elsif (!$g_fail_subonly) {
1078 foreach my $unresponsive (@g_unresponsive) {
1079 if (($_->[0] == $unresponsive->[0]) && !defined($_->[9]) && ($g_failover_method eq 'new')) {
1080 $line_prefix = "# (Node $_->[0] unavailable subscriber only) ";
1084 print SLONFILE ("# Preamble for node $_->[0] named $_->[4]\n");
1086 print SLONFILE ($line_prefix . "DEFINE $_->[4] $_->[0];\n");
1087 print SLONFILE ($line_prefix . "DEFINE $_->[4]_conninfo '$_->[2]';\n");
1088 print SLONFILE ($line_prefix . "NODE \@$_->[4] ADMIN CONNINFO = \@$_->[4]_conninfo;\n\n");
1091 print SLONFILE ($line_prefix . "NODE $_->[0] ADMIN CONNINFO = '$_->[2]';\n\n");
1093 if (($aliases) && ($sets)) {
1094 $set_count = loadSets($dbconninfo, $clname, $_->[0], $dbuser, $dbpass, $prefix, $logfile, $log_prefix);
1095 if ($set_count > 0) {
1096 print SLONFILE ("# Sets provided (currently) by node $_->[0]\n");
1097 foreach my $set (@g_sets) {
1098 print SLONFILE ($line_prefix . "DEFINE $set->[1] $set->[0];\n");
1100 print SLONFILE ("\n");
1107 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_write_fail', "script", $filename));
1113 # Write slonik commands to move sets
1116 my $dbconninfo = shift;
1123 my $aliases = shift;
1124 my $logfile = shift;
1125 my $log_prefix = shift;
1130 my $try_prefix = "";
1131 my ($year, $month, $day, $hour, $min, $sec) = (localtime(time))[5,4,3,2,1,0];
1132 my $filetime = sprintf ("%02d_%02d_%04d_%02d:%02d:%02d", $day, $month+1, $year+1900, $hour, $min, $sec);
1133 my $filename = $prefix . "/" . $clname . "-move_sets_from_" . $from . "_to_" . $to . "_on_" . $filetime . ".scr";
1135 if ($g_use_try_blocks) {
1142 my ($node, $setlist);
1143 my ($node_name, $setlist_name);
1146 unless (writePreamble($filename, $dbconninfo, $clname, $dbuser, $dbpass, true, $aliases, $prefix, $logfile, $log_prefix, true)) {
1147 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_incomplete_preamble'));
1150 foreach (@g_cluster) {
1151 if ($_->[0] == $from) {
1152 $from_name = $_->[4];
1154 elsif ($_->[0] == $to) {
1159 if (open(SLONFILE, ">>", $filename)) {
1161 print SLONFILE ("######\n# Actions (changes to cluster structure)\n######\n");
1163 $set_count = loadSets($dbconninfo, $clname, $from, $dbuser, $dbpass, $prefix, $logfile, $log_prefix);
1164 if ($set_count > 0) {
1166 if ($g_lockset_method ne "single") {
1167 if ($g_use_try_blocks) {
1168 print SLONFILE ("TRY {\n");
1172 print SLONFILE ($try_prefix . "ECHO 'Locking set $_->[1] ($_->[0])';\n");
1173 print SLONFILE ($try_prefix . "LOCK SET ( ID = \@$_->[1], ORIGIN = \@$from_name);\n");
1176 print SLONFILE ($try_prefix . "ECHO 'Locking set $_->[0]';\n");
1177 print SLONFILE ($try_prefix . "LOCK SET ( ID = $_->[0], ORIGIN = $from);\n");
1181 print SLONFILE ("\n");
1184 print SLONFILE ($try_prefix . "ECHO 'Moving set $_->[1] ($_->[0])';\n");
1185 print SLONFILE ($try_prefix . "MOVE SET ( ID = \@$_->[1], OLD ORIGIN = \@$from_name, NEW ORIGIN = \@$to_name);\n");
1188 print SLONFILE ($try_prefix . "ECHO 'Moving set $_->[0]';\n");
1189 print SLONFILE ($try_prefix . "MOVE SET ( ID = $_->[0], OLD ORIGIN = $from, NEW ORIGIN = $to);\n");
1193 if ($g_use_try_blocks) {
1194 print SLONFILE ("}\nON ERROR {\n");
1197 print SLONFILE ($try_prefix . "ECHO 'Unlocking set $_->[1] ($_->[0])';\n");
1198 print SLONFILE ($try_prefix . "UNLOCK SET ( ID = \@$_->[1], ORIGIN = \@$from_name);\n");
1201 print SLONFILE ($try_prefix . "ECHO 'Unlocking set $_->[0]';\n");
1202 print SLONFILE ($try_prefix . "UNLOCK SET ( ID = $_->[0], ORIGIN = $from);\n");
1205 print SLONFILE ("\tEXIT 1;\n}\nON SUCCESS {\n");
1208 print SLONFILE ($try_prefix . "WAIT FOR EVENT (ORIGIN = \@$from_name, CONFIRMED = ALL, WAIT ON = \@$from_name, TIMEOUT = 0);\n");
1211 print SLONFILE ($try_prefix . "WAIT FOR EVENT (ORIGIN = $from, CONFIRMED = ALL, WAIT ON = $from, TIMEOUT = 0);\n");
1213 if ($g_use_try_blocks) {
1214 print SLONFILE ("}\n");
1218 if ($g_lockset_method eq "single") {
1220 print SLONFILE ("\nECHO 'Moving set $_->[1] ($_->[0])';\n");
1221 if ($g_use_try_blocks) {
1222 print SLONFILE ("TRY {\n");
1224 print SLONFILE ($try_prefix . "LOCK SET ( ID = \@$_->[1], ORIGIN = \@$from_name);\n");
1225 print SLONFILE ($try_prefix . "MOVE SET ( ID = \@$_->[1], OLD ORIGIN = \@$from_name, NEW ORIGIN = \@$to_name);\n");
1226 if ($g_use_try_blocks) {
1227 print SLONFILE ("}\nON ERROR {\n" . $try_prefix . "UNLOCK SET ( ID = \@$_->[1], ORIGIN = \@$from_name);\n" . $try_prefix . "EXIT 1;\n}\n");
1229 print SLONFILE ("WAIT FOR EVENT (ORIGIN = \@$from_name, CONFIRMED = ALL, WAIT ON = \@$from_name, TIMEOUT = 0);\n");
1232 print SLONFILE ("\nECHO 'Moving set $_->[0]';\n");
1233 if ($g_use_try_blocks) {
1234 print SLONFILE ("TRY {\n");
1236 print SLONFILE ($try_prefix . "LOCK SET ( ID = $_->[0], ORIGIN = $from);\n");
1237 print SLONFILE ($try_prefix . "MOVE SET ( ID = $_->[0], OLD ORIGIN = $from, NEW ORIGIN = $to);\n");
1238 if ($g_use_try_blocks) {
1239 print SLONFILE ("}\nON ERROR {\n" . $try_prefix . "UNLOCK SET ( ID = $_->[0], ORIGIN = $from);\n" . $try_prefix . "EXIT 1;\n}\n");
1241 print SLONFILE ("WAIT FOR EVENT (ORIGIN = $from, CONFIRMED = ALL, WAIT ON = $from, TIMEOUT = 0);\n");
1244 if (($subs) && ($g_resubscribe_method eq 'subscribe')) {
1246 foreach my $other_subs (@g_cluster) {
1247 if (($other_subs->[6] eq "ACTIVE") && ($other_subs->[0] != $from) && ($other_subs->[0] != $to)) {
1249 if (exists $g_backups{$other_subs->[0]}) {
1250 $line_prefix = "# (Node $other_subs->[0] unavailable) ";
1256 # mess here needs cleaning up
1257 @subprov_name = (split(';', $other_subs->[8]));
1259 foreach $subprov (split(';', $other_subs->[5])) {
1260 ($node, $setlist) = (split('->', $subprov)) ;
1261 ($node_name, $setlist_name) = (split('->', $subprov_name[$subprov_idx])) ;
1264 $setlist =~ s/(\)|\(|s)//g;
1265 @subsets = (split(',', $setlist)) ;
1268 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_resubscribe', $_->[1], $_->[0], $other_subs->[0], $other_subs->[4], $setlist, $setlist_name, $node, $node_name));
1271 if ($_->[0] ~~ @subsets) {
1272 if ($node == $from) {
1274 print SLONFILE ($line_prefix .
1275 "ECHO 'Issuing subscribe for set $_->[1] ($_->[0]) provider $to_name ($to) -> " .
1276 "receiver $other_subs->[4] ($other_subs->[0])';\n");
1277 print SLONFILE ($line_prefix .
1278 "SUBSCRIBE SET ( ID = \@$_->[1], PROVIDER = \@$to_name, " .
1279 "RECEIVER = \@$other_subs->[4], FORWARD = YES);\n");
1282 print SLONFILE ($line_prefix .
1283 "ECHO 'Issuing subscribe for set $_->[1] ($_->[0]) provider $to -> " .
1284 "receiver $other_subs->[0]';\n");
1285 print SLONFILE ($line_prefix . "SUBSCRIBE SET ( ID = $_->[0], PROVIDER = $to, " .
1286 "RECEIVER = $other_subs->[0], FORWARD = YES);\n");
1291 print SLONFILE ($line_prefix .
1292 "ECHO 'Issuing subscribe for set $_->[1] ($_->[0]) provider $node_name ($node) -> " .
1293 "receiver $other_subs->[4] ($other_subs->[0])';\n");
1294 print SLONFILE ($line_prefix . "SUBSCRIBE SET ( ID = \@$_->[1], PROVIDER = \@$node_name, " .
1295 "RECEIVER = \@$other_subs->[4], FORWARD = YES);\n");
1298 print SLONFILE ($line_prefix .
1299 "ECHO 'Issuing subscribe for set $_->[1] ($_->[0]) provider $node -> " .
1300 "receiver $other_subs->[0]';\n");
1301 print SLONFILE ($line_prefix . "SUBSCRIBE SET ( ID = $_->[0], PROVIDER = $node, " .
1302 "RECEIVER = $other_subs->[0], FORWARD = YES);\n");
1312 if (($subs) && ($g_resubscribe_method eq 'resubscribe')) {
1314 foreach my $other_subs (@g_cluster) {
1315 if (($other_subs->[6] eq "ACTIVE") && ($other_subs->[0] != $from) && ($other_subs->[0] != $to)) {
1316 if (exists $g_backups{$other_subs->[0]}) {
1317 $line_prefix = "# (Node $other_subs->[0] unavailable) ";
1323 @subprov_name = (split(';', $other_subs->[8]));
1325 foreach $subprov (split(';', $other_subs->[5])) {
1326 ($node, $setlist) = (split('->', $subprov)) ;
1327 ($node_name, $setlist_name) = (split('->', $subprov_name[$subprov_idx])) ;
1331 print SLONFILE ("\n");
1332 if ($node == $from) {
1334 print SLONFILE ($line_prefix .
1335 "ECHO 'Issuing resubscribe for provider $to_name ($to) -> receiver $other_subs->[4] ($other_subs->[0])';\n");
1336 print SLONFILE ($line_prefix .
1337 "RESUBSCRIBE NODE ( ORIGIN = \@$to_name, PROVIDER = \@$to_name, RECEIVER = \@$other_subs->[4]);\n");
1340 print SLONFILE ($line_prefix .
1341 "ECHO 'Issuing resubscribe for provider $to -> receiver $other_subs->[0]';\n");
1342 print SLONFILE ($line_prefix .
1343 "RESUBSCRIBE NODE ( ORIGIN = $to, PROVIDER = $to, RECEIVER = $other_subs->[0] );\n");
1348 print SLONFILE ($line_prefix .
1349 "ECHO 'Issuing resubscribe for provider $node_name ($node) -> receiver $other_subs->[4] ($other_subs->[0])';\n");
1350 print SLONFILE ($line_prefix .
1351 "RESUBSCRIBE NODE ( ORIGIN = \@$to_name, PROVIDER = \@$node_name, RECEIVER = \@$other_subs->[4]);\n");
1354 print SLONFILE ($line_prefix .
1355 "ECHO 'Issuing resubscribe for provider $node -> receiver $other_subs->[0]';\n");
1356 print SLONFILE ($line_prefix .
1357 "RESUBSCRIBE NODE ( ORIGIN = $to, PROVIDER = $node, RECEIVER = $other_subs->[0]);\n");
1368 print SLONFILE ("\nECHO 'All sets originating from $from_name (id $from) have been moved to $to_name (id $to), ensure you modify any existing slonik scripts to reflect the new origin';\n");
1371 print SLONFILE ("\nECHO 'All sets originating from node $from have been moved to node $to, ensure you modify the any existing slonik scripts to reflect the new origin';\n");
1376 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_write_fail', "script", $filename));
1381 # Write slonik commands to failover sets
1384 my $dbconninfo = shift;
1391 my $aliases = shift;
1392 my $logfile = shift;
1393 my $log_prefix = shift;
1397 my ($year, $month, $day, $hour, $min, $sec) = (localtime(time))[5,4,3,2,1,0];
1398 my $filetime = sprintf ("%02d_%02d_%04d_%02d:%02d:%02d", $day, $month+1, $year+1900, $hour, $min, $sec);
1403 my ($node, $setlist);
1404 my ($node_name, $setlist_name);
1410 if (defined($from) && defined($to)) {
1411 $filename = $prefix . "/" . $clname . "-failover_from_" . $from . "_to_" . $to . "_on_" . $filetime . ".scr";
1414 $filename = $prefix . "/" . $clname . "-autofailover_on_" . $filetime . ".scr";
1417 if ($g_failover_method ne 'new') {
1418 # For pre 2.2 failover with multiple nodes, we attempt to resubscribe sets and drop other failed providers;
1419 # This will never work as well as 2.2+ failover behaviour (infact failover may not work as all in 2.0/2.1 with multiple failed nodes)
1420 # We also need to define the sets in the preamble for this.
1424 unless (writePreamble($filename, $dbconninfo, $clname, $dbuser, $dbpass, $sets, $aliases, $prefix, $logfile, $log_prefix, false)) {
1425 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_incomplete_preamble'));
1428 if (open(SLONFILE, ">>", $filename)) {
1430 print SLONFILE ("######\n# Actions (changes to cluster structure)\n######\n\n");
1432 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_failover_method',$g_failover_method));
1435 # If we are on pre 2.2 we need to drop failed subscriber nodes first regardless
1436 if ($g_failover_method ne 'new') {
1437 foreach (@g_failed) {
1438 if (!defined($_->[3])) {
1439 foreach my $backup (@g_cluster) {
1440 if ($backup->[0] == $g_backups{$_->[0]}) { # this backup node candidate is in the list of suitable nodes for {failed node}
1441 foreach my $subscriber (@g_cluster) {
1442 if (defined($subscriber->[1]) && $subscriber->[1] == $_->[0] && $subscriber->[0] != $backup->[0]) {
1443 # mess here needs cleaning up
1444 @subprov_name = (split(';', $subscriber->[8]));
1446 foreach my $subprov (split(';', $subscriber->[5])) {
1447 ($node, $setlist) = (split('->', $subprov)) ;
1448 ($node_name, $setlist_name) = (split('->', $subprov_name[$subprov_idx])) ;
1452 if ($node == $_->[0]) {
1454 print SLONFILE ("ECHO 'Resubscribing all sets on receiver $subscriber->[4] provided by other failed node $_->[4] to backup node $backup->[4]';\n");
1457 print SLONFILE ("ECHO 'Resubscribing all sets on receiver $subscriber->[0] provided by other failed node $_->[0] to backup node $backup->[0]';\n");
1459 $setlist =~ s/(\)|\(|s)//g;
1460 @subsets = (split(',', $setlist));
1461 $setlist_name =~ s/(\)|\()//g;
1462 @subsets_name = (split(',', $setlist_name));
1465 foreach my $subset (@subsets) {
1467 print SLONFILE ("SUBSCRIBE SET (ID = \@$subsets_name[$set_idx], PROVIDER = \@$backup->[4], RECEIVER = \@$subscriber->[4], FORWARD = YES);\n");
1468 print SLONFILE ("WAIT FOR EVENT (ORIGIN = \@$backup->[4], CONFIRMED = \@$subscriber->[4], WAIT ON = \@$backup->[4]);\n");
1471 print SLONFILE ("SUBSCRIBE SET (ID = $subset, PROVIDER = $backup->[0], RECEIVER = $subscriber->[0], FORWARD = YES);\n");
1472 print SLONFILE ("WAIT FOR EVENT (ORIGIN = $backup->[0], CONFIRMED = $subscriber->[0], WAIT ON = $backup->[0]);\n");
1476 print SLONFILE ("\n");
1481 print SLONFILE ("ECHO 'Dropping other failed node $_->[4] ($_->[0])';\n");
1482 print SLONFILE ("DROP NODE (ID = \@$_->[4], EVENT NODE = \@$backup->[4]);\n\n");
1485 print SLONFILE ("ECHO 'Dropping other failed node $_->[0]';\n");
1486 print SLONFILE ("DROP NODE (ID = $_->[0], EVENT NODE = $backup->[0]);\n\n");
1488 push(@dropped, $_->[0]);
1491 # The node is failed, but there are no downstream subscribers
1501 foreach (@g_failed) {
1502 if (($g_failover_method eq 'new') || defined($_->[3])) {
1503 foreach my $backup (@g_cluster) {
1504 if ($backup->[0] == $g_backups{$_->[0]}) {
1505 ## Here we have both details of the backup node and the failed node
1507 print SLONFILE ("ECHO 'Failing over slony cluster from $_->[4] (id $_->[0]) to $backup->[4] (id $backup->[0])';\n");
1510 print SLONFILE ("ECHO 'Failing over slony cluster from node $_->[0] to node $backup->[0]';\n");
1518 print SLONFILE ("FAILOVER (\n\t");
1520 foreach (@g_failed) {
1521 if (($g_failover_method eq 'new') || defined($_->[3])) {
1522 foreach my $backup (@g_cluster) {
1523 if ($backup->[0] == $g_backups{$_->[0]}) {
1524 ## Here we have both details of the backup node and the failed node
1525 if ($g_failover_method eq 'new') {
1526 if( $written != 0 ) {
1527 print SLONFILE (",\n\t");
1529 print SLONFILE ("NODE = (");
1532 if( $written != 0 ) {
1533 print SLONFILE ("\n);\nFAILOVER (\n\t");
1537 print SLONFILE ("ID = \@$_->[4], BACKUP NODE = \@$backup->[4]");
1540 print SLONFILE ("ID = $_->[0], BACKUP NODE = $backup->[0]");
1542 if ($g_failover_method eq 'new') {
1543 print SLONFILE (")");
1551 print SLONFILE ("\n);\n\n");
1553 if ($g_drop_failed) {
1554 if (($g_failover_method eq 'new') && (scalar(@g_failed) > 1)) {
1555 foreach (@g_failed) {
1557 print SLONFILE ("ECHO 'Dropping failed node $_->[4] ($_->[0])';\n");
1560 print SLONFILE ("ECHO 'Dropping failed node $_->[0]';\n");
1564 print SLONFILE ("DROP NODE (ID = '");
1568 foreach (@g_failed) {
1569 foreach my $backup (@g_cluster) {
1570 if ($backup->[0] == $g_backups{$_->[0]}) {
1571 if (!defined($event_node)) {
1573 $event_node = $backup->[4];
1576 $event_node = $backup->[0];
1579 if (($g_failover_method eq 'new') && (scalar(@g_failed) > 1)) {
1580 if( $written != 0 ) {
1581 print SLONFILE (",");
1583 ## Don't bother trying to define array values
1585 # print SLONFILE "\@$_->[4]";
1588 print SLONFILE $_->[0];
1592 elsif (($g_failover_method eq 'new') || defined($_->[3]) || !($_->[0] ~~ @dropped)) {
1594 print SLONFILE ("ECHO 'Dropping failed node $_->[4] ($_->[0])';\n");
1595 print SLONFILE ("DROP NODE (ID = \@$_->[4], EVENT NODE = \@$backup->[4]);\n\n");
1598 print SLONFILE ("ECHO 'Dropping failed node $_->[0]';\n");
1599 print SLONFILE ("DROP NODE (ID = $_->[0], EVENT NODE = $backup->[0]);\n\n");
1606 if (($g_failover_method eq 'new') && (scalar(@g_failed) > 1)) {
1608 print SLONFILE ("', EVENT NODE = \@$event_node);\n");
1611 print SLONFILE ("', EVENT NODE = $event_node);\n");
1618 printlog($prefix,$logfile,$log_prefix,lookupMsg('err_write_fail', "script", $filename));
1624 # Used to return informational text from the %message hashes, pretty much entirely stolen from check_postgres (http://bucardo.org)
1626 my $name = shift || '?';
1630 if (exists $message{$g_lang}{$name}) {
1631 $text = $message{$g_lang}{$name};
1633 elsif (exists $message{'en'}{$name}) {
1634 $text = $message{'en'}{$name};
1637 $line_call = (caller)[2];
1638 $text = qq{Failed to lookup text "$name" at line $line_call};
1644 $val = '?' if ! defined $val;
1645 last unless $text =~ s/\$$x/$val/g;
1652 # Trim quotes off a string
1655 $string =~ s/^('|")+//;
1656 $string =~ s/('|")+$//;
1663 $string =~ s/^\s+//;
1664 $string =~ s/\s+$//;
1668 # Print command with a linefeed
1670 print ((@_ ? join($/, @_) : $_), $/);
1673 # Print to stdout and the logfile, doing some replacements allong the way for logging
1676 my $logfile_name = shift;
1677 my $log_prefix = shift;
1678 my $message = shift;
1684 if (defined($logfile_name)) {
1686 # Do we have to do this all the time? Perhaps could check parameters first
1687 if ($logfile_name =~ /^\//i) {
1688 $logfile = strftime($logfile_name, localtime);
1691 $logfile = "$prefix/" . strftime($logfile_name, localtime);
1694 if ($log_prefix =~ m/(\%[mt])/) {
1695 my ($year, $month, $day, $hour, $min, $sec) = (localtime(time))[5,4,3,2,1,0];
1696 my ($h_sec, $h_msec) = gettimeofday;
1697 $date = sprintf ("%02d-%02d-%04d %02d:%02d:%02d.%03d", $day, $month+1, $year+1900, $hour, $min, $sec, $h_msec/1000);
1698 $log_prefix =~ s/\%m/$date/g;
1700 $date = sprintf ("%02d-%02d-%04d %02d:%02d:%02d", $day, $month+1, $year+1900, $hour, $min, $sec);
1701 $log_prefix =~ s/\%t/$date/g;
1703 if ($log_prefix =~ m/(\%p)/) {
1704 $log_prefix =~ s/\%p/$g_pid/g;
1707 if (open(LOGFILE, ">>", $logfile)) {
1708 print LOGFILE $log_prefix . " " . $message;
1712 println(lookupMsg('err_write_fail', "logfile", $logfile));
1717 # Printlog command with a linefeed
1719 printlog ($_[0], $_[1], $_[2], $_[3] . $/);
1722 # Insert details of any action into a database table
1724 my $dbconninfo = shift;
1727 my $exit_code = shift;
1730 my $logfile = shift;
1731 my $log_prefix = shift;
1743 unless($results = (read_file($logfile))) {
1744 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_read_fail', "logfile", $logfile));
1747 if (defined($script) && (-e $script)) {
1748 unless ($script_data = (read_file($script))) {
1749 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_read_fail', "script file", $script));
1753 $script_data = "No script data was generated.";
1754 $script = "No script generated.";
1757 $dsn = "DBI:Pg:$dbconninfo;";
1759 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
1760 $query = "INSERT INTO public.failovers (reason, exit_code, results, script, cluster_name)
1761 VALUES (?, ?, ?, ?, ?)";
1763 $sth = $dbh->prepare($query);
1765 $sth->bind_param(1, $reason);
1766 $sth->bind_param(2, $exit_code);
1767 $sth->bind_param(3, $results);
1768 $sth->bind_param(4, $script . ":\n" . $script_data);
1769 $sth->bind_param(5, $clname);
1778 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
1780 die lookupMsg('err_pgsql_connect');
1786 # Returns a uuid used for the failover script directory
1788 my $date_string = shift;
1789 my $g_ug = new Data::UUID;
1790 my $g_uuid = $g_ug->create_from_name("failover_script", $date_string);
1791 my $g_uuid_str = $g_ug->to_string($g_uuid);
1795 # Write out a PID file
1798 my $logfile = shift;
1799 my $log_prefix = shift;
1800 my $pidfile_name = shift;
1804 if ($pidfile_name =~ /^\//i) {
1805 $pidfile = $pidfile_name;
1808 $pidfile = "$prefix/" . $pidfile_name;
1811 open (PIDFILE, ">", $pidfile);
1817 printlogln($prefix,$logfile,$log_prefix, lookupMsg('dbg_generic', $!));
1819 printlogln($prefix,$logfile,$log_prefix, lookupMsg('err_write_fail', "pid file", $pidfile));
1825 # Remove the PID file
1828 my $logfile = shift;
1829 my $log_prefix = shift;
1830 my $pidfile_name = shift;
1834 if ($pidfile_name =~ /^\//i) {
1835 $pidfile = $pidfile_name;
1838 $pidfile = "$prefix/" . $pidfile_name;
1845 printlogln($prefix,$logfile,$log_prefix, lookupMsg('dbg_generic', 'PID file never existed to be removed'));
1850 printlogln($prefix,$logfile,$log_prefix, lookupMsg('dbg_generic', $!));
1852 printlogln($prefix,$logfile,$log_prefix, lookupMsg('err_unlink_fail', "pid file", $pidfile));
1858 # Check all sets from an originationg node are contained in the list provided by another node
1859 sub checkProvidesAllSets {
1860 my ($originSets, $providerSets) = @_;
1863 undef @test_hash{@$originSets}; # add a hash key for each element of @$originSets
1864 delete @test_hash{@$providerSets}; # remove all keys for elements of @$providerSets
1866 return !%test_hash; # return false if any keys are left in the hash
1869 # Check any sets from an originationg node are contained in the list subscribed to by another node
1870 sub checkSubscribesAnySets {
1871 my ($originSets, $subscriberSets) = @_;
1876 undef @test_hash{@$originSets}; # add a hash key for each element of @$originSets
1877 $before = scalar(keys %test_hash);
1878 delete @test_hash{@$subscriberSets}; # remove all keys for elements of @$subscriberSets
1879 $after = scalar(keys %test_hash);
1880 return ($before != $after); # return false if no keys were removed from the hash
1883 # Read configuration details from a configuration file
1885 my $cfgfile = shift;
1887 my $success = false;
1890 if (open(CFGFILE, "<", $cfgfile)) {
1891 foreach (<CFGFILE>) {
1896 s/#(?=(?:(?:[^']|[^"]*+'){2})*+[^']|[^"]*+\z).*//;
1898 if (length(trim($_))) {
1899 @fields = split('=', $_, 2);
1900 given(lc($fields[0])) {
1901 $value = qtrim(trim($fields[1]));
1905 when(/\bslony_database_host\b/i) {
1908 when(/\bslony_database_port\b/i) {
1909 $g_dbport = checkInteger($value);
1911 when(/\bslony_database_name\b/i) {
1914 when(/\bslony_database_user\b/i) {
1917 when(/\bslony_database_password\b/i) {
1920 when(/\bslony_cluster_name\b/i) {
1923 when(/\benable_debugging\b/i) {
1924 $g_debug = checkBoolean($value);
1926 when(/\bprefix_directory\b/i) {
1929 when(/\bseparate_working_directory\b/i) {
1930 $g_separate_working = checkBoolean($value);
1932 when(/\bpid_filename\b/i) {
1933 $g_pidfile = $value;
1935 when(/\bfailover_offline_subscriber_only\b/i) {
1936 $g_fail_subonly = checkBoolean($value);
1938 when(/\bdrop_failed_nodes\b/i) {
1939 $g_drop_failed = checkBoolean($value);
1941 when(/\blog_line_prefix\b/i) {
1942 $g_log_prefix = $value;
1944 when(/\blog_filename\b/i) {
1945 $g_logfile = $value;
1947 when(/\blog_to_postgresql\b/i) {
1948 $g_log_to_db = checkBoolean($value);
1950 when(/\blog_database_host\b/i) {
1951 $g_logdb_host = $value;
1953 when(/\blog_database_port\b/i) {
1954 $g_logdb_port = checkInteger($value);
1956 when(/\blog_database_name\b/i) {
1957 $g_logdb_name = $value;
1959 when(/\blog_database_user\b/i) {
1960 $g_logdb_user = $value;
1962 when(/\blog_database_password\b/i) {
1963 $g_logdb_pass = $value;
1965 when(/\benable_try_blocks\b/i) {
1966 $g_use_try_blocks = checkBoolean($value);
1968 when(/\bpull_aliases_from_comments\b/i) {
1969 $g_use_comment_aliases = checkBoolean($value);
1971 when(/\bslonik_path\b/i) {
1972 $g_slonikpath = $value;
1974 when(/\blockset_method\b/i) {
1975 $g_lockset_method = $value;
1977 when(/\benable_autofailover\b/i) {
1978 $g_autofailover = checkBoolean($value);
1980 when(/\bautofailover_poll_interval\b/i) {
1981 $g_autofailover_poll_interval = checkInteger($value);
1983 when(/\bautofailover_node_retry\b/i) {
1984 $g_autofailover_retry = checkInteger($value);
1986 when(/\bautofailover_sleep_time\b/i) {
1987 $g_autofailover_retry_sleep = checkInteger($value);
1989 when(/\bautofailover_forwarding_providers\b/i) {
1990 $g_autofailover_provs = checkBoolean($value);
1992 when(/\bautofailover_config_any_node\b/i) {
1993 $g_autofailover_config_any = checkBoolean($value);
1995 when(/\bautofailover_perspective_sleep_time\b/i) {
1996 $g_autofailover_perspective_sleep = checkInteger($value);
1998 when(/\bautofailover_majority_only\b/i) {
1999 $g_autofailover_majority_only = checkBoolean($value);
2001 when(/\bautofailover_is_quorum\b/i) {
2002 $g_autofailover_is_quorum = checkBoolean($value);
2012 println(lookupMsg('err_fail_config'));
2018 # Interpret a textual representation of a boolean value
2022 if ( grep /^$text$/i, ("y","yes","t","true","on") ) {
2025 elsif ( grep /^$text$/i, ("n","no","f","false","off") ) {
2031 # Check if a text value is a valid integer
2033 my $integer = shift;
2036 if (($integer * 1) eq $integer) {
2037 $value = int($integer);
2042 # Run a slonik command and capture all output via autoflushing channel
2046 my $logfile = shift;
2047 my $log_prefix = shift;
2051 printlogln($prefix,$logfile,$log_prefix, lookupMsg('dbg_slonik_script', $script));
2053 if (open(SLONIKSTATUS, "-|", "slonik $script 2>&1")) {
2054 while (<SLONIKSTATUS>) {
2055 printlogln($prefix,$logfile,$log_prefix,lookupMsg('slonik_output', $_));
2057 close(SLONIKSTATUS);
2061 printlogln($prefix,$logfile,$log_prefix, lookupMsg('err_running_slonik', $!));
2067 # Experimental logic to watch the cluster status and perform an automatic failover
2069 my $dbconninfo = shift;
2074 my $logfile = shift;
2075 my $log_prefix = shift;
2082 my $cluster_loaded_from;
2087 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_init'));
2088 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_init_cnf', ($g_autofailover_config_any ? 'any' : 'specified target')));
2089 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_init_pol', $g_autofailover_poll_interval));
2090 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_init_ret', $g_autofailover_retry, $g_autofailover_retry_sleep));
2091 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_init_set', ($g_autofailover_provs ? 'will' : 'will not')));
2094 # Probe current cluster configuration every minute
2095 if (!defined($cluster_time) || (time()-$cluster_time > 60)) {
2097 $cluster_loaded = false;
2098 if (!defined($cluster_time) || !$g_autofailover_config_any) {
2100 ($node_count, $version) = loadCluster($dbconninfo, $clname, $dbuser, $dbpass, $prefix, $logfile, $log_prefix);
2101 die lookupMsg('err_cluster_empty') if ($node_count == 0);
2102 @cluster = @g_cluster;
2103 die lookupMsg('err_cluster_lone') if ($node_count == 1);
2104 $cluster_loaded = true;
2105 $cluster_loaded_from = 'conninfo specified in config';
2108 printlogln($prefix,$logfile,$log_prefix, lookupMsg('load_cluster_fail', 'from supplied configuration'));
2110 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
2115 foreach (@cluster) {
2116 if ($_->[6] eq "ACTIVE") {
2117 unless ($cluster_loaded) {
2119 ($node_count, $version) = loadCluster($_->[2], $clname, $dbuser, $dbpass, $prefix, $logfile, $log_prefix);
2120 die lookupMsg('err_cluster_empty') if ($node_count == 0);
2121 @cluster = @g_cluster;
2122 die lookupMsg('err_cluster_lone') if ($node_count == 1);
2123 $cluster_loaded = true;
2124 $cluster_loaded_from = $_->[0];
2127 printlogln($prefix,$logfile,$log_prefix, lookupMsg('load_cluster_fail', 'from node ' . $_->[0] . ': trying next node'));
2129 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
2137 if ($cluster_loaded) {
2138 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_load_cluster', (!defined($cluster_time) ? "Loaded" : "Reloaded"), $version, $clname, $node_count, $cluster_loaded_from));
2139 $cluster_time = time();
2142 printlogln($prefix,$logfile,$log_prefix, lookupMsg('load_cluster_fail', 'from any node'));
2146 if ($cluster_loaded) {
2149 while(($current_retry <= $g_autofailover_retry) && ((!defined($failed)) || ($failed > 0))) {
2150 # Check status of cluster
2151 $failed = checkFailed($clname, $dbuser, $dbpass, $prefix, $logfile, $log_prefix);
2154 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_cluster_good'));
2156 if ($current_retry > 0) {
2157 printlogln($prefix,$logfile,$log_prefix,lookupMsg('cluster_fixed'));
2161 if (($failed > 0) && ($current_retry <= $g_autofailover_retry)) {
2162 printlogln($prefix,$logfile,$log_prefix,lookupMsg('cluster_failed', $failed,$g_autofailover_retry_sleep,$current_retry,$g_autofailover_retry));
2163 usleep($g_autofailover_retry_sleep * 1000);
2167 if ((!$g_autofailover_majority_only || checkSplit($prefix, $logfile, $log_prefix)) && (($g_autofailover_perspective_sleep <= 0) || checkPerspective($clname, $dbuser, $dbpass, $prefix, $logfile, $log_prefix))) {
2168 $actions = findBackup($clname, $dbuser, $dbpass, $prefix, $logfile, $log_prefix);
2170 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_proceed'));
2171 foreach my $failed ( keys %g_backups ) {
2172 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_detail', $failed, $g_backups{$failed}));
2174 $g_script = writeFailover($prefix, $dbconninfo, $clname, $dbuser, $dbpass, undef, undef, $g_subs_follow_origin, $g_use_comment_aliases, $logfile, $log_prefix);
2175 unless (runSlonik($g_script, $prefix, $logfile, $log_prefix)) {
2176 printlogln($prefix,$logfile,$log_prefix,lookupMsg('err_execute_fail', 'slonik script', $g_script));
2178 $cluster_loaded = false;
2180 #print "SCRIPT: $g_script\n";
2184 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_halt', $failed));
2188 usleep($g_autofailover_poll_interval * 1000);
2197 # Count of failed and live nodes to perform very basic split-brain check
2200 my $logfile = shift;
2201 my $log_prefix = shift;
2203 my $majority = false;
2204 my $failed = scalar(@g_unresponsive);
2205 my $survivers = (scalar(@g_cluster) - scalar(@g_unresponsive));
2207 if ($survivers > $failed) {
2209 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_split_check', $survivers, ($survivers+$failed)));
2211 elsif (($survivers == $failed) && $g_autofailover_is_quorum) {
2213 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_split_check', ($survivers . '+quorum'), ($survivers+$failed)));
2216 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_split_check_fail', $survivers));
2222 # Check each nodes perspective of the failure to try to ensure the issue isn't that this script just can't connect to the origin/provider
2223 # The idea here is just to wait for a short period of time and see if the lag time for the nodes has increased by the same amount
2224 sub checkPerspective {
2229 my $logfile = shift;
2230 my $log_prefix = shift;
2239 my @unresponsive_ids;
2246 foreach (@g_unresponsive) {
2247 push(@unresponsive_ids, $_->[0]);
2249 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check', join(", ", @unresponsive_ids), scalar(@g_unresponsive), scalar(@g_cluster)));
2251 foreach (@g_cluster) {
2252 unless ($_->[0] ~~ @unresponsive_ids) {
2253 $dsn = "DBI:Pg:$_->[2];";
2255 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
2256 $qw_clname = $dbh->quote_identifier("_" . $clname);
2258 $query = "SELECT a.st_origin, a.st_received, extract(epoch from a.st_lag_time)::integer
2259 FROM _test_replication.sl_status a
2260 INNER JOIN _test_replication.sl_node b on a.st_origin = b.no_id
2261 INNER JOIN _test_replication.sl_node c on a.st_received = c.no_id
2262 WHERE a.st_received IN (" . substr('?, ' x scalar(@unresponsive_ids), 0, -2) . ") ORDER BY a.st_origin, a.st_received;";
2264 $sth = $dbh->prepare($query);
2267 foreach (@unresponsive_ids) {
2268 $sth->bind_param($param_on, $_);
2273 while (my @node_lag = $sth->fetchrow) {
2274 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check_data', 'Check1', $_->[0], $node_lag[0], $node_lag[1], $node_lag[2]));
2275 push(@lag_info1, \@node_lag);
2282 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check_fail', $_->[0], $@));
2289 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check_sleep', $g_autofailover_perspective_sleep));
2290 usleep($g_autofailover_perspective_sleep * 1000);
2292 foreach (@g_cluster) {
2293 unless ($_->[0] ~~ @unresponsive_ids) {
2294 $dsn = "DBI:Pg:$_->[2];";
2296 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
2297 $qw_clname = $dbh->quote_identifier("_" . $clname);
2299 $query = "SELECT a.st_origin, a.st_received, extract(epoch from a.st_lag_time)::integer
2300 FROM _test_replication.sl_status a
2301 INNER JOIN _test_replication.sl_node b on a.st_origin = b.no_id
2302 INNER JOIN _test_replication.sl_node c on a.st_received = c.no_id
2303 WHERE a.st_received IN (" . substr('?, ' x scalar(@unresponsive_ids), 0, -2) . ") ORDER BY a.st_origin, a.st_received;";
2305 $sth = $dbh->prepare($query);
2308 foreach (@unresponsive_ids) {
2309 $sth->bind_param($param_on, $_);
2314 while (my @node_lag = $sth->fetchrow) {
2315 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check_data', 'Check2', $_->[0], $node_lag[0], $node_lag[1], $node_lag[2]));
2316 push(@lag_info2, \@node_lag);
2323 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check_fail', $_->[0], $@));
2331 foreach (@lag_info1) {
2333 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', ("Node $_->[0] lag between checks on node $_->[1] is " . ($lag_info2[$lag_idx]->[2]-$_->[2]) . " seconds")));
2336 if ((($lag_info2[$lag_idx]->[2]-$_->[2])*1000) >= $g_autofailover_perspective_sleep) {
2344 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check_unknown'));
2346 elsif ($lag_idx == $lag_confirmed) {
2347 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check_true'));
2351 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_pspec_check_false'));
2357 # Check if any nodes have failed by connecting and probing the Slony schema
2363 my $logfile = shift;
2364 my $log_prefix = shift;
2370 my $result_count = 0;
2371 my $prov_failed = 0;
2372 my $subonly_failed = 0;
2374 undef @g_unresponsive;
2376 foreach (@g_cluster) {
2377 if ($_->[6] eq "ACTIVE") {
2379 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_autofailover_check',$_->[0], ($_->[4] // "unnamed"),(defined($_->[9]) ? "provider of sets $_->[9]" : "sole subscriber"),$_->[2]));
2383 if ((defined($_->[3])) || ($g_autofailover_provs && defined($_->[9]))) {
2384 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_autofailover_active_check', 'provider', $_->[0]));
2387 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_autofailover_active_check', 'subscriber only', $_->[0]));
2391 $dsn = "DBI:Pg:$_->[2];";
2393 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
2394 $query = "SELECT count(*) FROM pg_namespace WHERE nspname = ?";
2395 $sth = $dbh->prepare($query);
2396 $sth->bind_param(1, "_" . $clname);
2399 $result_count = $result_count+$sth->rows;
2404 if (exists($g_unresponsive_subonly{$_->[0]})) {
2405 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_recovery_subonly', $_->[0]));
2406 delete $g_unresponsive_subonly{$_->[0]};
2411 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
2413 push(@g_unresponsive, \@$_);
2414 if ((defined($_->[3])) || ($g_autofailover_provs && defined($_->[9]))) {
2415 if (defined($_->[3])) {
2416 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_unresponsive', $_->[0]));
2419 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_unresponsive_prov', $_->[0]));
2421 unless ($g_failover_method ne 'new' && !defined($_->[3])) {
2426 if (!exists($g_unresponsive_subonly{$_->[0]})) {
2427 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_unresponsive_subonly', $_->[0]));
2428 $g_unresponsive_subonly{$_->[0]} = true;
2430 if ($g_fail_subonly) {
2438 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_autofailover_check',$_->[0], ($_->[4] // "unnamed"), lc($_->[6] // "unknown") . ' node', $_->[2]));
2442 if ($prov_failed > 0) {
2443 return ($prov_failed+$subonly_failed);
2446 return $prov_failed;
2450 # Attempt to try and find the most suitable backup node for a failed node
2456 my $logfile = shift;
2457 my $log_prefix = shift;
2464 my $result_count = 0;
2465 my $lowest_lag_time;
2466 my $latest_last_event;
2468 my $best_node_is_direct;
2469 my $best_node_can_forward;
2473 my %backup_for_set_chosen;
2478 foreach (@g_unresponsive) {
2479 if ($g_fail_subonly || (defined($_->[3])) || ($g_autofailover_provs && defined($_->[9]))) {
2480 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_promote_find', ($_->[9] // "none"), $_->[0]));
2482 undef $best_node_id;
2483 $lowest_lag_time = (1<<$Config{ivsize}*8-1)-1;
2484 $latest_last_event = 0;
2485 $best_node_is_direct = false;
2486 $best_node_can_forward = false;
2488 if (defined($_->[9]) && (exists $backup_for_set_chosen{$_->[9]})) {
2489 $best_node_id = $backup_for_set_chosen{$_->[9]};
2490 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_promote_found', $_->[9], $_->[0]));
2493 foreach my $subscriber (@g_cluster) {
2494 if ($subscriber->[0] != $_->[0]) {
2496 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_check_sub',$subscriber->[0]));
2499 # Here the strings containing the sets are converted to arrays
2500 # Origin / Forwarded sets
2501 undef @sets_to_prov;
2502 if (defined($subscriber->[10])) {
2503 @sets_to_prov = split(',',$subscriber->[10]);
2505 if (defined($subscriber->[3])) {
2506 if (@sets_to_prov) {
2507 @sets_to_prov = (@sets_to_prov, split(',',$subscriber->[3]));
2510 @sets_to_prov = split(',',$subscriber->[3]);
2513 if (!defined($subscriber->[10]) && !defined($subscriber->[3])) {
2514 @sets_to_prov = (0);
2517 # Origin / Subscribed sets.
2519 if (defined($subscriber->[7])) {
2520 @sets_to = split(',',$subscriber->[7]);
2522 if (defined($subscriber->[3])) {
2524 @sets_to = (@sets_to, split(',',$subscriber->[3]));
2527 @sets_to = split(',',$subscriber->[3]);
2530 if (!defined($subscriber->[7]) && !defined($subscriber->[3])) {
2534 # Sets provided by the failed node.
2536 if (defined($_->[9])) {
2537 @sets_from = split(',',$_->[9]);
2544 $dsn = "DBI:Pg:$subscriber->[2]";
2547 $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
2548 $qw_clname = $dbh->quote_identifier("_" . $clname);
2550 #print "\tNODE " . $subscriber->[0] . ") SETS TO = " . join(',', @sets_to) . " SETS FROM = " . join(',', @sets_from) . " SETS TO PROV = " . join(',', @sets_to_prov) . "\n";
2552 if (defined($subscriber->[3]) && checkProvidesAllSets(\@sets_from, \@sets_to)) {
2553 $query = "SELECT 0, ev_seqno, (ev_origin = ?)
2554 FROM $qw_clname.sl_event
2555 WHERE ev_origin = $qw_clname.getlocalnodeid(?)
2556 ORDER BY ev_seqno DESC LIMIT 1";
2559 $query = "SELECT extract(epoch from (current_timestamp-a.con_timestamp)), a.con_seqno, (a.con_origin = ?) AS direct
2560 FROM $qw_clname.sl_confirm a
2561 INNER JOIN $qw_clname.sl_event b on b.ev_seqno = a.con_seqno AND a.con_origin = b.ev_origin
2562 INNER JOIN $qw_clname.sl_subscribe c ON c.sub_provider = a.con_origin AND c.sub_receiver = a.con_received
2563 WHERE c.sub_active AND a.con_received = $qw_clname.getlocalnodeid(?)
2564 ORDER BY a.con_seqno DESC LIMIT 1;";
2567 $sth = $dbh->prepare($query);
2568 $sth->bind_param(1, $_->[0]);
2569 $sth->bind_param(2, "_" . $clname);
2572 while (my @subinfo = $sth->fetchrow) {
2573 # If the failed provider node isn't an origin for any sets, we classify any direct subscribers to it as indeirect
2574 # because they are indirect to the origin.
2575 if ($subinfo[2] && defined($_->[9]) && !defined($_->[3])) {
2576 $subinfo[2] = false;
2579 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_node_detail',
2581 ($subinfo[2]?"directly":"indirectly"),
2582 (defined($_->[3])?"origin":(defined($_->[9])?"provider":"subscriber only")),
2584 (defined($subscriber->[10])?$subscriber->[10]:(defined($subscriber->[3])?$subscriber->[3]:"<NONE>")),
2585 $subinfo[0], $subinfo[1]));
2587 # If select this node as the backup node if:
2588 # 1) The node is a subscriber to all sets on the failed node
2589 # 2) In order of preference:
2590 # The node is one of the direct subscribers to the failed node on the most recent event and is a forwarding provider
2592 # The node is one of the direct subscribers to the failed node on the most recent event and is not a forwarding provider
2594 # The node is an indirect subscriber to the failed node with the lowest lag time
2595 if (!checkProvidesAllSets(\@sets_from, \@sets_to)) {
2596 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_promote_unsuitable', $subscriber->[0]));
2598 elsif (($subinfo[2] && (!$best_node_is_direct || $subinfo[1] > $latest_last_event || (!$best_node_can_forward && checkProvidesAllSets(\@sets_from, \@sets_to_prov) && $subinfo[1] == $latest_last_event)))
2599 || (!$best_node_is_direct && !$subinfo[2] && $subinfo[0] < $lowest_lag_time)) {
2600 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_promote_best', $subscriber->[0], $subinfo[0], $subinfo[1]));
2601 $best_node_id = $subscriber->[0];
2602 $lowest_lag_time = $subinfo[0];
2603 $latest_last_event = $subinfo[1];
2604 $best_node_is_direct = $subinfo[2];
2605 $best_node_can_forward = checkProvidesAllSets(\@sets_from, \@sets_to_prov);
2611 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_check_sub_fail', $subscriber->[0]));
2613 printlogln($prefix,$logfile,$log_prefix,lookupMsg('dbg_generic', $@));
2619 if (defined($best_node_id)) {
2620 push(@g_failed, \@$_);
2621 $g_backups{$_->[0]} = $best_node_id;
2622 if (defined($_->[9]) && !(exists $g_backups{$_->[9]})) {
2623 $backup_for_set_chosen{$_->[9]} = $best_node_id;
2627 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_promote_fail'));
2631 printlogln($prefix,$logfile,$log_prefix,lookupMsg('autofailover_promote_skip', $_->[0]));
2634 return keys(%g_backups);