/*
* Test-Connection: This is called from the threaded context inside the
* dispatcher: ONLY CALL THREADSAFE FUNCTIONS -- elog() is NOT threadsafe.
*/
bool
FtsTestConnection(CdbComponentDatabaseInfo *failedDBInfo, bool fullScan)
{
/* master is always reported as alive */
if (failedDBInfo->segindex == MASTER_SEGMENT_ID)
{
return true;
}
if (!fullScan)
{
return FTS_STATUS_ISALIVE(failedDBInfo->dbid, ftsProbeInfo->fts_status);
}
FtsNotifyProber();
return FTS_STATUS_ISALIVE(failedDBInfo->dbid, ftsProbeInfo->fts_status);
}
/*
* Build a set of changes, based on our current state, and the probe results.
*/
static bool
probePublishUpdate(uint8 *probe_results)
{
bool update_found = false;
int i;
if (failover_strategy == 'f')
{
/* preprocess probe results to decide what is the current segment state */
FtsPreprocessProbeResultsFilerep(cdb_component_dbs, probe_results);
}
for (i = 0; i < cdb_component_dbs->total_segment_dbs; i++)
{
CdbComponentDatabaseInfo *segInfo = &cdb_component_dbs->segment_db_info[i];
/* if we've gotten a pause or shutdown request, we ignore our probe results. */
if (!FtsIsActive())
{
return false;
}
/* we check segments in pairs of primary-mirror */
if (!SEGMENT_IS_ACTIVE_PRIMARY(segInfo))
{
continue;
}
CdbComponentDatabaseInfo *primary = segInfo;
CdbComponentDatabaseInfo *mirror = FtsGetPeerSegment(segInfo->segindex, segInfo->dbid);
if (failover_strategy == 'n')
{
Assert(SEGMENT_IS_ACTIVE_PRIMARY(segInfo));
Assert(FTS_STATUS_ISALIVE(segInfo->dbid, ftsProbeInfo->fts_status));
Assert(mirror == NULL);
/* no mirror available to failover */
if (!PROBE_IS_ALIVE(segInfo))
{
FtsSegmentStatusChange changes;
uint8 statusOld = ftsProbeInfo->fts_status[segInfo->dbid];
uint8 statusNew = statusOld & ~FTS_STATUS_ALIVE;
buildSegmentStateChange(segInfo, &changes, statusNew);
FtsFailoverNull(&changes);
}
continue;
}
Assert(failover_strategy == 'f' || failover_strategy == 's');
Assert(mirror != NULL);
/* changes required for primary and mirror */
FtsSegmentStatusChange changes[2];
uint32 stateOld = 0;
uint32 stateNew = 0;
bool isPrimaryAlive = PROBE_IS_ALIVE(primary);
bool isMirrorAlive = PROBE_IS_ALIVE(mirror);
/* get transition type */
uint32 trans = getTransition(isPrimaryAlive, isMirrorAlive);
if (gp_log_fts > GPVARS_VERBOSITY_VERBOSE)
{
elog(LOG, "FTS: primary found %s, mirror found %s, transition %d.",
(isPrimaryAlive ? "alive" : "dead"), (isMirrorAlive ? "alive" : "dead"), trans);
}
if (trans == TRANS_D_D)
{
elog(LOG, "FTS: detected double failure for content=%d, primary (dbid=%d), mirror (dbid=%d).",
primary->segindex, primary->dbid, mirror->dbid);
}
if (failover_strategy == 'f')
{
/* get current state */
stateOld = FtsGetPairStateFilerep(primary, mirror);
/* get new state */
stateNew = transition(stateOld, trans, primary, mirror, &changes[0], &changes[1]);
}
else
{
Assert(failover_strategy == 's');
/* get current state */
stateOld = FtsGetPairStateSAN(primary, mirror);
/* get new state */
stateNew = transition(stateOld, trans, primary, mirror, &changes[0], &changes[1]);
}
/* check if transition is required */
if (stateNew != stateOld)
{
update_found = true;
updateConfiguration(changes, ARRAY_SIZE(changes));
}
}
if (gp_log_fts >= GPVARS_VERBOSITY_VERBOSE)
{
elog(LOG, "FTS: probe result processing is complete.");
}
return update_found;
}
/**
* Marks the given db as in-sync in the segment configuration.
*/
void
FtsMarkSegmentsInSync(CdbComponentDatabaseInfo *primary, CdbComponentDatabaseInfo *mirror)
{
if (!FTS_STATUS_ISALIVE(primary->dbid, ftsProbeInfo->fts_status) ||
!FTS_STATUS_ISALIVE(mirror->dbid, ftsProbeInfo->fts_status) ||
!FTS_STATUS_ISPRIMARY(primary->dbid, ftsProbeInfo->fts_status) ||
FTS_STATUS_ISPRIMARY(mirror->dbid, ftsProbeInfo->fts_status) ||
FTS_STATUS_IS_SYNCED(primary->dbid, ftsProbeInfo->fts_status) ||
FTS_STATUS_IS_SYNCED(mirror->dbid, ftsProbeInfo->fts_status) ||
FTS_STATUS_IS_CHANGELOGGING(primary->dbid, ftsProbeInfo->fts_status) ||
FTS_STATUS_IS_CHANGELOGGING(mirror->dbid, ftsProbeInfo->fts_status))
{
FtsRequestPostmasterShutdown(primary, mirror);
}
if (ftsProbeInfo->fts_pauseProbes)
{
return;
}
uint8 segStatus=0;
Relation configrel;
Relation histrel;
ScanKeyData scankey;
SysScanDesc sscan;
HeapTuple configtuple;
HeapTuple newtuple;
HeapTuple histtuple;
Datum configvals[Natts_gp_segment_configuration];
bool confignulls[Natts_gp_segment_configuration] = { false };
bool repls[Natts_gp_segment_configuration] = { false };
Datum histvals[Natts_gp_configuration_history];
bool histnulls[Natts_gp_configuration_history] = { false };
char *desc = "FTS: changed segment to insync from resync.";
/*
* Commit/abort transaction below will destroy
* CurrentResourceOwner. We need it for catalog reads.
*/
ResourceOwner save = CurrentResourceOwner;
StartTransactionCommand();
/* update primary */
segStatus = ftsProbeInfo->fts_status[primary->dbid];
segStatus |= FTS_STATUS_SYNCHRONIZED;
ftsProbeInfo->fts_status[primary->dbid] = segStatus;
/* update mirror */
segStatus = ftsProbeInfo->fts_status[mirror->dbid];
segStatus |= FTS_STATUS_SYNCHRONIZED;
ftsProbeInfo->fts_status[mirror->dbid] = segStatus;
histrel = heap_open(GpConfigHistoryRelationId,
RowExclusiveLock);
configrel = heap_open(GpSegmentConfigRelationId,
RowExclusiveLock);
/* update gp_segment_configuration to insync */
ScanKeyInit(&scankey,
Anum_gp_segment_configuration_dbid,
BTEqualStrategyNumber, F_INT2EQ,
Int16GetDatum(primary->dbid));
sscan = systable_beginscan(configrel, GpSegmentConfigDbidIndexId,
true, SnapshotNow, 1, &scankey);
configtuple = systable_getnext(sscan);
if (!HeapTupleIsValid(configtuple))
{
elog(ERROR,"FTS cannot find dbid (%d, %d) in %s", primary->dbid,
mirror->dbid, RelationGetRelationName(configrel));
}
configvals[Anum_gp_segment_configuration_mode-1] = CharGetDatum('s');
repls[Anum_gp_segment_configuration_mode-1] = true;
newtuple = heap_modify_tuple(configtuple, RelationGetDescr(configrel),
configvals, confignulls, repls);
simple_heap_update(configrel, &configtuple->t_self, newtuple);
CatalogUpdateIndexes(configrel, newtuple);
systable_endscan(sscan);
ScanKeyInit(&scankey,
Anum_gp_segment_configuration_dbid,
BTEqualStrategyNumber, F_INT2EQ,
Int16GetDatum(mirror->dbid));
sscan = systable_beginscan(configrel, GpSegmentConfigDbidIndexId,
true, SnapshotNow, 1, &scankey);
configtuple = systable_getnext(sscan);
if (!HeapTupleIsValid(configtuple))
{
elog(ERROR,"FTS cannot find dbid (%d, %d) in %s", primary->dbid,
mirror->dbid, RelationGetRelationName(configrel));
}
newtuple = heap_modify_tuple(configtuple, RelationGetDescr(configrel),
configvals, confignulls, repls);
simple_heap_update(configrel, &configtuple->t_self, newtuple);
CatalogUpdateIndexes(configrel, newtuple);
systable_endscan(sscan);
/* update configuration history */
histvals[Anum_gp_configuration_history_time-1] =
TimestampTzGetDatum(GetCurrentTimestamp());
histvals[Anum_gp_configuration_history_dbid-1] =
Int16GetDatum(primary->dbid);
histvals[Anum_gp_configuration_history_desc-1] =
CStringGetTextDatum(desc);
histtuple = heap_form_tuple(RelationGetDescr(histrel), histvals, histnulls);
simple_heap_insert(histrel, histtuple);
CatalogUpdateIndexes(histrel, histtuple);
histvals[Anum_gp_configuration_history_dbid-1] =
Int16GetDatum(mirror->dbid);
histtuple = heap_form_tuple(RelationGetDescr(histrel), histvals, histnulls);
simple_heap_insert(histrel, histtuple);
CatalogUpdateIndexes(histrel, histtuple);
ereport(LOG,
(errmsg("FTS: resynchronization of mirror (dbid=%d, content=%d) on %s:%d has completed.",
mirror->dbid, mirror->segindex, mirror->address, mirror->port ),
errSendAlert(true)));
heap_close(histrel, RowExclusiveLock);
heap_close(configrel, RowExclusiveLock);
/*
* Do not block shutdown. We will always get a change to update
* gp_segment_configuration in subsequent probes upon database
* restart.
*/
if (shutdown_requested)
{
elog(LOG, "Shutdown in progress, ignoring FTS prober updates.");
return;
}
CommitTransactionCommand();
CurrentResourceOwner = save;
}
