database/openGauss-server
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

!846 修复一主多备下备机故障时主机walsender超时未退出的问题

Browse Source 
Merge pull request !846 from chenxiaobin/walsnd_timeout
This commit is contained in:
opengauss-bot
2021-03-22 20:14:29 +08:00
committed by Gitee
parent c3e4163462 5b25cfb87b
commit 79c44ca48d
3 changed files with 12 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/gausskernel/process/threadpool/knl_thread.cpp
View File

@ -1301,6 +1301,7 @@ static void knl_t_walsender_init(knl_t_walsender_context* walsender_cxt)
walsender_cxt->tmpbuf = (StringInfoData*)palloc0(sizeof(StringInfoData));
walsender_cxt->remotePort = 0;
walsender_cxt->walSndCaughtUp = false;
walsender_cxt->last_check_timeout_timestamp = 0;
}
static void knl_t_tsearch_init(knl_t_tsearch_context* tsearch_cxt)

18
src/gausskernel/storage/replication/walsender.cpp
View File

@ -3499,26 +3499,26 @@ static void WalSndCheckTimeOut(TimestampTz now)
}
/*
* Use static last_reply_time to avoid call GetHeartbeatLastReplyTimestamp frequently
* Use last_check_timeout_timestamp to avoid call GetHeartbeatLastReplyTimestamp frequently
* when t_thrd.walsender_cxt.last_reply_timestamp has meet the timeout condition
* but last heartbeat time doesn't.
*/
static TimestampTz last_reply_time = t_thrd.walsender_cxt.last_reply_timestamp;
TimestampTz *last_reply_time = &t_thrd.walsender_cxt.last_check_timeout_timestamp;
/* t_thrd.walsender_cxt.last_reply_timestamp newer */
if (timestamptz_cmp_internal(t_thrd.walsender_cxt.last_reply_timestamp, last_reply_time) > 0) {
last_reply_time = t_thrd.walsender_cxt.last_reply_timestamp;
if (timestamptz_cmp_internal(t_thrd.walsender_cxt.last_reply_timestamp, *last_reply_time) > 0) {
*last_reply_time = t_thrd.walsender_cxt.last_reply_timestamp;
}
timeout = CalculateTimeout(last_reply_time);
timeout = CalculateTimeout(*last_reply_time);
if (now < timeout) {
return;
}
TimestampTz heartbeat = GetHeartbeatLastReplyTimestamp();
/* If heartbeat newer, use heartbeat to recalculate timeout. */
if (timestamptz_cmp_internal(heartbeat, last_reply_time) > 0) {
last_reply_time = heartbeat;
timeout = CalculateTimeout(last_reply_time);
if (timestamptz_cmp_internal(heartbeat, *last_reply_time) > 0) {
*last_reply_time = heartbeat;
timeout = CalculateTimeout(*last_reply_time);
}
if (now >= timeout) {
@ -3529,7 +3529,7 @@ static void WalSndCheckTimeOut(TimestampTz now)
*/
if (log_min_messages <= ERROR || client_min_messages <= ERROR) {
WalReplicationTimestampInfo timeStampInfo;
WalReplicationTimestampToString(&timeStampInfo, now, timeout, last_reply_time, heartbeat);
WalReplicationTimestampToString(&timeStampInfo, now, timeout, *last_reply_time, heartbeat);
ereport(ERROR, (errmsg("terminating Walsender process due to replication timeout."
"now time(%s) timeout time(%s) last recv time(%s) heartbeat time(%s)",
timeStampInfo.nowTimeStamp, timeStampInfo.timeoutStamp,

2
src/include/knl/knl_thread.h
View File

@ -2189,6 +2189,8 @@ typedef struct knl_t_walsender_context {
bool walSndCaughtUp;
/* Notify primary to advance logical replication slot. */
struct pg_conn* advancePrimaryConn;
/* Timestamp of the last check-timeout time in WalSndCheckTimeOut. */
TimestampTz last_check_timeout_timestamp;
} knl_t_walsender_context;
typedef struct knl_t_walreceiverfuncs_context {