repmgr: improve slot handling in "node rejoin"

On the rejoined node, if a replication slot for the new upstream exists
(which is typically the case after a failover), delete that slot.

Also emit a warning about any inactive replication slots which may need
to be cleaned up manually.

GitHub #499.
This commit is contained in:
Ian Barwick
2018-08-30 11:18:08 +09:00
parent e468ca859e
commit 3fb20ce774
4 changed files with 98 additions and 44 deletions

View File

@@ -2978,3 +2978,46 @@ can_use_pg_rewind(PGconn *conn, const char *data_directory, PQExpBufferData *rea
return can_use;
}
void
drop_replication_slot_if_exists(PGconn *conn, int node_id, char *slot_name)
{
t_replication_slot slot_info = T_REPLICATION_SLOT_INITIALIZER;
RecordStatus record_status = get_slot_record(conn, slot_name, &slot_info);
log_verbose(LOG_DEBUG, "attempting to delete slot \"%s\" on node %i",
slot_name, node_id);
if (record_status != RECORD_FOUND)
{
/* this is a good thing */
log_verbose(LOG_INFO,
_("slot \"%s\" does not exist on node %i, nothing to remove"),
slot_name, node_id);
}
else
{
if (slot_info.active == false)
{
if (drop_replication_slot(conn, slot_name) == true)
{
log_notice(_("replication slot \"%s\" deleted on node %i"), slot_name, node_id);
}
else
{
log_error(_("unable to delete replication slot \"%s\" on node %i"), slot_name, node_id);
}
}
/*
* if active replication slot exists, call Houston as we have a
* problem
*/
else
{
log_warning(_("replication slot \"%s\" is still active on node %i"), slot_name, node_id);
}
}
}