(cloudberry) 15/17: Confine Greenplum-specific WAL replication behavior to coordinator

[yj...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

yjhjstz pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/cloudberry.git

commit 6039d4ed4ff8b199318e8ad286b7e157ca60b27c
Author: Asim R P <pa...@vmware.com>
AuthorDate: Tue May 18 21:58:06 2021 +0530

    Confine Greenplum-specific WAL replication behavior to coordinator
    
    Although WAL replication is used as the HA mechanism by all segments
    in a Greenplum cluster, there are some differences in how it is used
    by coordinator/standby and by primary/mirror segments.  One such
    difference is the use of replication slots.  Primary/mirror WAL
    replication makes use of replication slots to ensure WAL that a
    primary retains WAL that may still be needed by its mirror.
    
    Replication slots are not used for coordinator/standby.  To ensure
    that coordinator retains WAL that is not yet received by the standby,
    possibly due to replication lag, a special mechanism is used by
    coordinator to track the flush location of standby.
    
    This patch confinces the use of this special mechanism to coordinator,
    the mechanism is no longer used by primary segments.  Ideally,
    coordinator/standby may also make use of replication slots.  The GUC
    max_slot_wal_keep_size can be set to the lowest value of
    (checkpoint_segments * XLOG_BLCKSZ) on coordinator to ensure that WAL
    doesn't fill up disk space when the standby goes down.  That could
    done in a followup patch.
    
    Reviewed by: Ashwin Agrawal <aa...@vmware.com>
    Reviewed by: Paul Guo <pa...@gmail.com>
    
    (cherry picked from commit 0f2d65f497245eb00b5fc544cf609e79ddd75226)
---
 src/backend/access/transam/xlog.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index df5ae6b3c3..79bfd61bf5 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -1013,7 +1013,6 @@ static int	get_sync_bit(int method);
 
 /* New functions added for WAL replication */
 static void XLogProcessCheckpointRecord(XLogReaderState *rec);
-
 static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch,
 								XLogRecData *rdata,
 								XLogRecPtr StartPos, XLogRecPtr EndPos);
@@ -10523,8 +10522,14 @@ KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo, XLogRecPtr PriorRedoPtr)
 	/*
 	 * Calculate how many segments are kept by slots first, adjusting for
 	 * max_slot_wal_keep_size.
+	 *
+	 * Greenplum: coordinator needs a different way to determine the keep
+	 * point as replication slot is not created there.
 	 */
-	keep = XLogGetReplicationSlotMinimumLSN();
+	keep = IS_QUERY_DISPATCHER() ?
+		WalSndCtlGetXLogCleanUpTo() :
+		XLogGetReplicationSlotMinimumLSN();
+
 #ifdef FAULT_INJECTOR
 	/*
 	 * Let the WAL still needed be removed.  This is used to test if WAL sender


---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cloudberry.apache.org
For additional commands, e-mail: commits-help@cloudberry.apache.org