github/repmgr
1
0
Fork 0
mirror of https://github.com/EnterpriseDB/repmgr.git synced 2026-03-22 22:56:29 +00:00
Code Issues Packages Projects Releases 1 Wiki Activity

Add %v, %u and %t parameters to "failover_validation_command"

Browse Source 
These indicate:
 - the number of visible nodes sharing the current upstream
 - the number of nodes on the current upstream
 - the total number of nodes in the entire repmgr cluster.

This allows the failover_validation_command to be used to perform
more thorough validations, including cross-referencing external
cluster management state (e.g. if managed by kubernetes).

GitHub #651.
This commit is contained in:
Stanislav Paskalev
2020-06-04 15:31:33 +03:00
committed by Ian Barwick
parent f1bdb09512
commit 73e8373337
3 changed files with 75 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
dbutils.c
View File

@@ -2907,6 +2907,37 @@ get_all_node_records(PGconn *conn, NodeInfoList *node_list)
return success;
}
bool
get_all_nodes_count(PGconn *conn, int *count)
{
PQExpBufferData query;
PGresult *res = NULL;
bool success = true;
initPQExpBuffer(&query);
appendPQExpBufferStr(&query,
" SELECT count(*) "
" FROM repmgr.nodes n ");
log_verbose(LOG_DEBUG, "get_all_nodes_count():\n%s", query.data);
res = PQexec(conn, query.data);
if (PQresultStatus(res) != PGRES_TUPLES_OK)
{
log_db_error(conn, query.data, _("get_all_nodes_count(): unable to execute query"));
success = false;
}
else
{
*count = atoi(PQgetvalue(res, 0, 0));
}
PQclear(res);
termPQExpBuffer(&query);
return success;
}
void
get_downstream_node_records(PGconn *conn, int node_id, NodeInfoList *node_list)

1
dbutils.h
View File

@@ -496,6 +496,7 @@ bool get_local_node_record(PGconn *conn, int node_id, t_node_info *node_info);
bool get_primary_node_record(PGconn *conn, t_node_info *node_info);
bool get_all_node_records(PGconn *conn, NodeInfoList *node_list);
bool get_all_nodes_count(PGconn *conn, int *count);
void get_downstream_node_records(PGconn *conn, int node_id, NodeInfoList *nodes);
void get_active_sibling_node_records(PGconn *conn, int node_id, int upstream_node_id, NodeInfoList *node_list);
bool get_child_nodes(PGconn *conn, int node_id, NodeInfoList *node_list);

58
repmgrd-physical.c
View File

@@ -52,6 +52,12 @@ typedef enum
ELECTION_RERUN
} ElectionResult;
typedef struct election_stats
{
int visible_nodes;
int shared_upstream_nodes;
int all_nodes;
} election_stats;
typedef struct t_child_node_info
{
@@ -114,8 +120,8 @@ static bool update_monitoring_history(void);
static void handle_sighup(PGconn **conn, t_server_type server_type);
static const char *format_failover_state(FailoverState failover_state);
static ElectionResult execute_failover_validation_command(t_node_info *node_info);
static void parse_failover_validation_command(const char *template, t_node_info *node_info, PQExpBufferData *out);
static ElectionResult execute_failover_validation_command(t_node_info *node_info, election_stats *stats);
static void parse_failover_validation_command(const char *template, t_node_info *node_info, election_stats *stats, PQExpBufferData *out);
static bool check_node_can_follow(PGconn *local_conn, XLogRecPtr local_xlogpos, PGconn *follow_target_conn, t_node_info *follow_target_node_info);
static void check_witness_attached(t_node_info *node_info, bool startup);
@@ -4093,13 +4099,10 @@ do_election(NodeInfoList *sibling_nodes, int *new_primary_id)
{
int electoral_term = -1;
/* we're visible */
int visible_nodes = 1;
int total_nodes = 0;
NodeInfoListCell *cell = NULL;
t_node_info *candidate_node = NULL;
election_stats stats;
ReplInfo local_replication_info;
@@ -4120,6 +4123,11 @@ do_election(NodeInfoList *sibling_nodes, int *new_primary_id)
int nodes_with_primary_still_visible = 0;
/* we're visible */
stats.visible_nodes = 1;
stats.shared_upstream_nodes = 0;
stats.all_nodes = 0;
electoral_term = get_current_term(local_conn);
if (electoral_term == -1)
@@ -4157,7 +4165,11 @@ do_election(NodeInfoList *sibling_nodes, int *new_primary_id)
log_info(_("%i active sibling nodes registered"), sibling_nodes->node_count);
total_nodes = sibling_nodes->node_count + 1;
stats.shared_upstream_nodes = sibling_nodes->node_count + 1;
get_all_nodes_count(local_conn, &stats.all_nodes);
log_info(_("%i total nodes registered"), stats.all_nodes);
if (strncmp(upstream_node_info.location, local_node_info.location, MAXLEN) != 0)
{
@@ -4184,7 +4196,7 @@ do_election(NodeInfoList *sibling_nodes, int *new_primary_id)
{
if (config_file_options.failover_validation_command[0] != '\0')
{
return execute_failover_validation_command(&local_node_info);
return execute_failover_validation_command(&local_node_info, &stats);
}
log_info(_("no other sibling nodes - we win by default"));
@@ -4280,7 +4292,7 @@ do_election(NodeInfoList *sibling_nodes, int *new_primary_id)
cell->node_info->node_status = NODE_STATUS_UP;
visible_nodes++;
stats.visible_nodes++;
/*
* see if the node is in the primary's location (but skip the check if
@@ -4562,11 +4574,11 @@ do_election(NodeInfoList *sibling_nodes, int *new_primary_id)
termPQExpBuffer(&nodes_with_primary_visible);
log_info(_("visible nodes: %i; total nodes: %i; no nodes have seen the primary within the last %i seconds"),
visible_nodes,
total_nodes,
stats.visible_nodes,
stats.shared_upstream_nodes,
(config_file_options.monitor_interval_secs * 2));
if (visible_nodes <= (total_nodes / 2.0))
if (stats.visible_nodes <= (stats.shared_upstream_nodes / 2.0))
{
log_notice(_("unable to reach a qualified majority of nodes"));
log_detail(_("node will enter degraded monitoring state waiting for reconnect"));
@@ -4592,7 +4604,7 @@ do_election(NodeInfoList *sibling_nodes, int *new_primary_id)
if (config_file_options.failover_validation_command[0] != '\0')
{
return execute_failover_validation_command(candidate_node);
return execute_failover_validation_command(candidate_node, &stats);
}
return ELECTION_WON;
@@ -4822,7 +4834,7 @@ handle_sighup(PGconn **conn, t_server_type server_type)
}
static ElectionResult
execute_failover_validation_command(t_node_info *node_info)
execute_failover_validation_command(t_node_info *node_info, election_stats *stats)
{
PQExpBufferData failover_validation_command;
PQExpBufferData command_output;
@@ -4833,6 +4845,7 @@ execute_failover_validation_command(t_node_info *node_info)
parse_failover_validation_command(config_file_options.failover_validation_command,
node_info,
stats,
&failover_validation_command);
log_notice(_("executing \"failover_validation_command\""));
@@ -4871,7 +4884,7 @@ execute_failover_validation_command(t_node_info *node_info)
static void
parse_failover_validation_command(const char *template, t_node_info *node_info, PQExpBufferData *out)
parse_failover_validation_command(const char *template, t_node_info *node_info, election_stats *stats, PQExpBufferData *out)
{
const char *src_ptr;
@@ -4896,6 +4909,21 @@ parse_failover_validation_command(const char *template, t_node_info *node_info,
src_ptr++;
appendPQExpBufferStr(out, node_info->node_name);
break;
case 'v':
/* %v: visible nodes count */
src_ptr++;
appendPQExpBuffer(out, "%i", stats->visible_nodes);
break;
case 'u':
/* %u: shared upstream nodes count */
src_ptr++;
appendPQExpBuffer(out, "%i", stats->shared_upstream_nodes);
break;
case 't':
/* %t: total nodes count */
src_ptr++;
appendPQExpBuffer(out, "%i", stats->all_nodes);
break;
default:
/* otherwise treat the % as not special */