/*
* Cleanup work if something breaks while executing the query
*/
void
Persistent_ExecuteQuery_Cleanup(void)
{
Assert(oldMemoryContext && savedResourceOwner);
if (connected)
{
SPI_finish();
connected=false;
}
AbortCurrentTransaction();
MemoryContextSwitchTo(oldMemoryContext);
CurrentResourceOwner = savedResourceOwner;
oldMemoryContext = NULL;
savedResourceOwner = NULL;
}
/*
* kafka_consume_main
*
* Main function for Kafka consumers running as background workers
*/
void
kafka_consume_main(Datum arg)
{
char err_msg[512];
rd_kafka_topic_conf_t *topic_conf;
rd_kafka_t *kafka;
rd_kafka_topic_t *topic;
rd_kafka_message_t **messages;
const struct rd_kafka_metadata *meta;
struct rd_kafka_metadata_topic topic_meta;
rd_kafka_resp_err_t err;
bool found;
Oid id = (Oid) arg;
ListCell *lc;
KafkaConsumerProc *proc = hash_search(consumer_procs, &id, HASH_FIND, &found);
KafkaConsumer consumer;
CopyStmt *copy;
int valid_brokers = 0;
int i;
int my_partitions = 0;
if (!found)
elog(ERROR, "kafka consumer %d not found", id);
pqsignal(SIGTERM, kafka_consume_main_sigterm);
#define BACKTRACE_SEGFAULTS
#ifdef BACKTRACE_SEGFAULTS
pqsignal(SIGSEGV, debug_segfault);
#endif
/* we're now ready to receive signals */
BackgroundWorkerUnblockSignals();
/* give this proc access to the database */
BackgroundWorkerInitializeConnection(NameStr(proc->dbname), NULL);
/* load saved consumer state */
StartTransactionCommand();
load_consumer_state(proc->consumer_id, &consumer);
copy = get_copy_statement(&consumer);
topic_conf = rd_kafka_topic_conf_new();
kafka = rd_kafka_new(RD_KAFKA_CONSUMER, NULL, err_msg, sizeof(err_msg));
rd_kafka_set_logger(kafka, logger);
/*
* Add all brokers currently in pipeline_kafka_brokers
*/
if (consumer.brokers == NIL)
elog(ERROR, "no valid brokers were found");
foreach(lc, consumer.brokers)
valid_brokers += rd_kafka_brokers_add(kafka, lfirst(lc));
if (!valid_brokers)
elog(ERROR, "no valid brokers were found");
/*
* Set up our topic to read from
*/
topic = rd_kafka_topic_new(kafka, consumer.topic, topic_conf);
err = rd_kafka_metadata(kafka, false, topic, &meta, CONSUMER_TIMEOUT);
if (err != RD_KAFKA_RESP_ERR_NO_ERROR)
elog(ERROR, "failed to acquire metadata: %s", rd_kafka_err2str(err));
Assert(meta->topic_cnt == 1);
topic_meta = meta->topics[0];
load_consumer_offsets(&consumer, &topic_meta, proc->offset);
CommitTransactionCommand();
/*
* Begin consuming all partitions that this process is responsible for
*/
for (i = 0; i < topic_meta.partition_cnt; i++)
{
int partition = topic_meta.partitions[i].id;
Assert(partition <= consumer.num_partitions);
if (partition % consumer.parallelism != proc->partition_group)
continue;
elog(LOG, "[kafka consumer] %s <- %s consuming partition %d from offset %ld",
consumer.rel->relname, consumer.topic, partition, consumer.offsets[partition]);
if (rd_kafka_consume_start(topic, partition, consumer.offsets[partition]) == -1)
elog(ERROR, "failed to start consuming: %s", rd_kafka_err2str(rd_kafka_errno2err(errno)));
my_partitions++;
}
/*
* No point doing anything if we don't have any partitions assigned to us
*/
if (my_partitions == 0)
{
elog(LOG, "[kafka consumer] %s <- %s consumer %d doesn't have any partitions to read from",
consumer.rel->relname, consumer.topic, MyProcPid);
goto done;
}
messages = palloc0(sizeof(rd_kafka_message_t) * consumer.batch_size);
/*
* Consume messages until we are terminated
*/
while (!got_sigterm)
{
ssize_t num_consumed;
int i;
int messages_buffered = 0;
int partition;
StringInfoData buf;
bool xact = false;
for (partition = 0; partition < consumer.num_partitions; partition++)
{
if (partition % consumer.parallelism != proc->partition_group)
continue;
num_consumed = rd_kafka_consume_batch(topic, partition,
CONSUMER_TIMEOUT, messages, consumer.batch_size);
if (num_consumed <= 0)
continue;
if (!xact)
{
StartTransactionCommand();
xact = true;
}
initStringInfo(&buf);
for (i = 0; i < num_consumed; i++)
{
if (messages[i]->payload != NULL)
{
appendBinaryStringInfo(&buf, messages[i]->payload, messages[i]->len);
if (buf.len > 0 && buf.data[buf.len - 1] != '\n')
appendStringInfoChar(&buf, '\n');
messages_buffered++;
}
consumer.offsets[partition] = messages[i]->offset;
rd_kafka_message_destroy(messages[i]);
}
}
if (!xact)
{
pg_usleep(1 * 1000);
continue;
}
/* we don't want to die in the event of any errors */
PG_TRY();
{
if (messages_buffered)
execute_copy(copy, &buf);
}
PG_CATCH();
{
elog(LOG, "[kafka consumer] %s <- %s failed to process batch, dropped %d message%s:",
consumer.rel->relname, consumer.topic, (int) num_consumed, (num_consumed == 1 ? "" : "s"));
EmitErrorReport();
FlushErrorState();
AbortCurrentTransaction();
xact = false;
}
PG_END_TRY();
if (!xact)
StartTransactionCommand();
if (messages_buffered)
save_consumer_state(&consumer, proc->partition_group);
CommitTransactionCommand();
}
done:
hash_search(consumer_procs, &id, HASH_REMOVE, NULL);
rd_kafka_topic_destroy(topic);
rd_kafka_destroy(kafka);
rd_kafka_wait_destroyed(CONSUMER_TIMEOUT);
}
static void
QDMirroringUpdate(
QDMIRRORUpdateMask updateMask,
bool validFlag,
QDMIRRORState state,
QDMIRRORDisabledReason disabledReason,
struct timeval *lastLogTimeVal,
char *errorMessage)
{
#define UPDATE_VALIDFLAG_CMD "update gp_configuration set valid='%c' where dbid = CAST(%d AS SMALLINT)"
#define UPDATE_MASTER_MIRRORING_CMD "update gp_master_mirroring set (summary_state, detail_state, log_time, error_message) = ('%s', %s, '%s'::timestamptz, %s);"
int count = 0;
char cmd[200 + QDMIRRORErrorMessageSize * 2 + 3];
char detailValue[100];
char logTimeStr[128];
char *summaryStateString;
char *detailStateString;
char errorMessageQuoted[QDMIRRORErrorMessageSize * 2 + 3];
char *user;
MemoryContext mcxt = CurrentMemoryContext;
Segment *master = NULL;
volatile PQExpBuffer entryBuffer = NULL;
volatile PGconn *entryConn = NULL;
volatile PGresult *rs = NULL;
PG_TRY();
{
StartTransactionCommand();
user = getDBSuperuserName("QDMirroringUpdate");
Assert(user != NULL);
master = GetMasterSegment();
entryBuffer = createPQExpBuffer();
if (PQExpBufferBroken(entryBuffer))
{
destroyPQExpBuffer(entryBuffer);
ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("QDMirroringUpdate: out of memory")));
/* not reached. */
}
/*
* initialize libpq connection buffer, we only need to initialize it
* once.
*/
initPQConnectionBuffer(master, user, NULL, entryBuffer, true);
FreeSegment(master);
free(user);
/*
* Call libpq to connect
*/
entryConn = PQconnectdb(entryBuffer->data);
if (PQstatus((PGconn *)entryConn) == CONNECTION_BAD)
{
/*
* When we get an error, we strdup it here. When the main thread
* checks for errors, it makes a palloc copy of this, and frees
* this.
*/
char *error_message = strdup(PQerrorMessage((PGconn *)entryConn));
if (!error_message)
{
ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("QDMirroringUpdate: out of memory")));
}
destroyPQExpBuffer(entryBuffer);
PQfinish((PGconn *)entryConn);
entryConn = NULL;
elog(FATAL, "QDMirroringUpdate: setting segDB state failed, error connecting to entry db, error: %s", error_message);
}
/* finally, we're ready to actually get some stuff done. */
do
{
rs = PQexec((PGconn *)entryConn, "BEGIN");
if (PQresultStatus((PGresult *)rs) != PGRES_COMMAND_OK)
break;
if ((updateMask & QDMIRROR_UPDATEMASK_VALIDFLAG) != 0)
{
count = snprintf(cmd, sizeof(cmd), UPDATE_VALIDFLAG_CMD, (validFlag ? 't' : 'f'), ftsQDMirrorInfo->dbid);
if (count >= sizeof(cmd))
{
ereport(ERROR, (errcode(ERRCODE_INTERNAL_ERROR),
errmsg("QDMirroringUpdate: format command string failure")));
}
rs = PQexec((PGconn *)entryConn, cmd);
if (PQresultStatus((PGresult *)rs) != PGRES_COMMAND_OK)
{
ereport(ERROR, (errcode(ERRCODE_INTERNAL_ERROR),
errmsg("QDMirroringUpdate: could not execute command '%s'", cmd)));
break;
}
}
if ((updateMask & QDMIRROR_UPDATEMASK_MASTERMIRRORING)!= 0)
{
switch (state)
{
case QDMIRROR_STATE_NONE:
summaryStateString = "None";
break;
case QDMIRROR_STATE_NOTCONFIGURED:
summaryStateString = "Not Configured";
break;
case QDMIRROR_STATE_CONNECTINGWALSENDSERVER:
case QDMIRROR_STATE_POSITIONINGTOEND:
case QDMIRROR_STATE_CATCHUPPENDING:
case QDMIRROR_STATE_CATCHINGUP:
summaryStateString = "Synchronizing";
break;
case QDMIRROR_STATE_SYNCHRONIZED:
summaryStateString = "Synchronized";
break;
case QDMIRROR_STATE_DISABLED:
summaryStateString = "Not Synchronized";
break;
default:
summaryStateString = "Unknown";
break;
}
if (state == QDMIRROR_STATE_DISABLED)
{
detailStateString =
QDMirroringDisabledReasonToString(disabledReason);
}
else
{
detailStateString = NULL;
}
if (detailStateString == NULL)
{
strcpy(detailValue, "null");
}
else
{
count = snprintf(detailValue, sizeof(detailValue), "'%s'", detailStateString);
if (count >= sizeof(detailValue))
{
ereport(ERROR, (errcode(ERRCODE_INTERNAL_ERROR),
errmsg("QDMirroringUpdate: format command string failure")));
}
}
QDMirroringFormatTime(logTimeStr, sizeof(logTimeStr), lastLogTimeVal);
/*
* Escape quote the error string before putting in DML statement...
*/
if (errorMessage != NULL)
{
int errorMessageLen = strlen(errorMessage);
if (errorMessageLen == 0)
{
strcpy(errorMessageQuoted, "null");
}
else
{
size_t escapedLen;
errorMessageQuoted[0] = '\'';
escapedLen = PQescapeString(&errorMessageQuoted[1], errorMessage, errorMessageLen);
errorMessageQuoted[escapedLen + 1] = '\'';
errorMessageQuoted[escapedLen + 2] = '\0';
elog((Debug_print_qd_mirroring ? LOG : DEBUG5), "Error message quoted: \"%s\"", errorMessageQuoted);
}
}
else
{
strcpy(errorMessageQuoted, "null");
}
count = snprintf(cmd, sizeof(cmd), UPDATE_MASTER_MIRRORING_CMD,
summaryStateString, detailValue, logTimeStr, errorMessageQuoted);
if (count >= sizeof(cmd))
{
ereport(ERROR, (errcode(ERRCODE_INTERNAL_ERROR),
errmsg("QDMirroringUpdate: format command string failure")));
}
rs = PQexec((PGconn *)entryConn, cmd);
if (PQresultStatus((PGresult *)rs) != PGRES_COMMAND_OK)
{
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("QDMirroringUpdate: could not execute command '%s'", cmd)));
break;
}
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),
"Successfully executed command \"%s\"", cmd);
rs = PQexec((PGconn *)entryConn, "COMMIT");
if (PQresultStatus((PGresult *)rs) != PGRES_COMMAND_OK)
break;
}
}
while (0);
PQclear((PGresult *)rs);
PQfinish((PGconn *)entryConn);
destroyPQExpBuffer(entryBuffer);
CommitTransactionCommand();
}
PG_CATCH();
{
PQclear((PGresult *)rs);
PQfinish((PGconn *)entryConn);
destroyPQExpBuffer(entryBuffer);
AbortCurrentTransaction();
}
PG_END_TRY();
MemoryContextSwitchTo(mcxt); /* Just incase we hit an error */
return;
}
void
ContQuerySchedulerMain(int argc, char *argv[])
{
sigjmp_buf local_sigjmp_buf;
List *dbs = NIL;
/* we are a postmaster subprocess now */
IsUnderPostmaster = true;
am_cont_scheduler = true;
/* reset MyProcPid */
MyProcPid = getpid();
MyPMChildSlot = AssignPostmasterChildSlot();
/* record Start Time for logging */
MyStartTime = time(NULL);
/* Identify myself via ps */
init_ps_display("continuous query scheduler process", "", "", "");
ereport(LOG, (errmsg("continuous query scheduler started")));
if (PostAuthDelay)
pg_usleep(PostAuthDelay * 1000000L);
SetProcessingMode(InitProcessing);
/*
* If possible, make this process a group leader, so that the postmaster
* can signal any child processes too. This is only for consistency sake, we
* never fork the scheduler process. Instead dynamic bgworkers are used.
*/
#ifdef HAVE_SETSID
if (setsid() < 0)
elog(FATAL, "setsid() failed: %m");
#endif
/*
* Set up signal handlers. We operate on databases much like a regular
* backend, so we use the same signal handling. See equivalent code in
* tcop/postgres.c.
*/
pqsignal(SIGHUP, sighup_handler);
pqsignal(SIGINT, sigint_handler);
pqsignal(SIGTERM, sigterm_handler);
pqsignal(SIGQUIT, quickdie);
InitializeTimeouts(); /* establishes SIGALRM handler */
pqsignal(SIGPIPE, SIG_IGN);
pqsignal(SIGUSR1, procsignal_sigusr1_handler);
pqsignal(SIGUSR2, sigusr2_handler);
pqsignal(SIGFPE, FloatExceptionHandler);
pqsignal(SIGCHLD, SIG_DFL);
#define BACKTRACE_SEGFAULTS
#ifdef BACKTRACE_SEGFAULTS
pqsignal(SIGSEGV, debug_segfault);
#endif
/* Early initialization */
BaseInit();
/*
* Create a per-backend PGPROC struct in shared memory, except in the
* EXEC_BACKEND case where this was done in SubPostmasterMain. We must do
* this before we can use LWLocks (and in the EXEC_BACKEND case we already
* had to do some stuff with LWLocks).
*/
#ifndef EXEC_BACKEND
InitProcess();
#endif
InitPostgres(NULL, InvalidOid, NULL, NULL);
SetProcessingMode(NormalProcessing);
/*
* Create a memory context that we will do all our work in. We do this so
* that we can reset the context during error recovery and thereby avoid
* possible memory leaks.
*/
ContQuerySchedulerMemCxt = AllocSetContextCreate(TopMemoryContext,
"ContQuerySchedulerCtx",
ALLOCSET_DEFAULT_MINSIZE,
ALLOCSET_DEFAULT_INITSIZE,
ALLOCSET_DEFAULT_MAXSIZE);
MemoryContextSwitchTo(ContQuerySchedulerMemCxt);
/*
* If an exception is encountered, processing resumes here.
*
* This code is a stripped down version of PostgresMain error recovery.
*/
if (sigsetjmp(local_sigjmp_buf, 1) != 0)
{
/* since not using PG_TRY, must reset error stack by hand */
error_context_stack = NULL;
/* Prevents interrupts while cleaning up */
HOLD_INTERRUPTS();
/* Forget any pending QueryCancel or timeout request */
disable_all_timeouts(false);
QueryCancelPending = false; /* second to avoid race condition */
/* Report the error to the server log */
EmitErrorReport();
/* Abort the current transaction in order to recover */
AbortCurrentTransaction();
/*
* Now return to normal top-level context and clear ErrorContext for
* next time.
*/
MemoryContextSwitchTo(ContQuerySchedulerMemCxt);
FlushErrorState();
/* Flush any leaked data in the top-level context */
MemoryContextResetAndDeleteChildren(ContQuerySchedulerMemCxt);
/* Now we can allow interrupts again */
RESUME_INTERRUPTS();
/*
* Sleep at least 1 second after any error. We don't want to be
* filling the error logs as fast as we can.
*/
pg_usleep(1000000L);
}
/* We can now handle ereport(ERROR) */
PG_exception_stack = &local_sigjmp_buf;
/* must unblock signals before calling rebuild_database_list */
PG_SETMASK(&UnBlockSig);
ContQuerySchedulerShmem->scheduler_pid = MyProcPid;
dbs = get_database_list();
/* Loop forever */
for (;;)
{
ListCell *lc;
int rc;
foreach(lc, dbs)
{
DatabaseEntry *db_entry = lfirst(lc);
bool found;
ContQueryProcGroup *grp = hash_search(ContQuerySchedulerShmem->proc_table, &db_entry->oid, HASH_ENTER, &found);
/* If we don't have an entry for this dboid, initialize a new one and fire off bg procs */
if (!found)
{
grp->db_oid = db_entry->oid;
namestrcpy(&grp->db_name, NameStr(db_entry->name));
start_group(grp);
}
}
/* Allow sinval catchup interrupts while sleeping */
EnableCatchupInterrupt();
/*
* Wait until naptime expires or we get some type of signal (all the
* signal handlers will wake us by calling SetLatch).
*/
rc = WaitLatch(&MyProc->procLatch, WL_LATCH_SET | WL_POSTMASTER_DEATH, 0);
ResetLatch(&MyProc->procLatch);
DisableCatchupInterrupt();
/*
* Emergency bailout if postmaster has died. This is to avoid the
* necessity for manual cleanup of all postmaster children.
*/
if (rc & WL_POSTMASTER_DEATH)
proc_exit(1);
/* the normal shutdown case */
if (got_SIGTERM)
break;
/* update config? */
if (got_SIGHUP)
{
got_SIGHUP = false;
ProcessConfigFile(PGC_SIGHUP);
/* update tuning parameters, so that they can be read downstream by background processes */
update_tuning_params();
}
/* terminate a proc group? */
if (got_SIGUSR2)
{
HASH_SEQ_STATUS status;
ContQueryProcGroup *grp;
got_SIGUSR2 = false;
hash_seq_init(&status, ContQuerySchedulerShmem->proc_table);
while ((grp = (ContQueryProcGroup *) hash_seq_search(&status)) != NULL)
{
ListCell *lc;
if (!grp->terminate)
continue;
foreach(lc, dbs)
{
DatabaseEntry *entry = lfirst(lc);
if (entry->oid == grp->db_oid)
{
dbs = list_delete(dbs, entry);
break;
}
}
terminate_group(grp);
}
}
