wait_event_type for WAIT FOR LSN
wait_event_type for WAIT FOR LSN
От:
Noah Misch <noah@leadboat.com>
Дата:
commit 7a39f43 wrote: > --- a/src/backend/utils/activity/wait_event_names.txt > +++ b/src/backend/utils/activity/wait_event_names.txt > @@ -76,34 +76,35 @@ ABI_compatibility: > # Wait Events - Client > # > # Use this category when a process is waiting to send data to or receive data > # from the frontend process to which it is connected. This is never used for > # a background process, which has no client connection. > # > > Section: ClassName - WaitEventClient > > CLIENT_READ "Waiting to read data from the client." > CLIENT_WRITE "Waiting to write data to the client." > GSS_OPEN_SERVER "Waiting to read data from the client while establishing a GSSAPI session." > LIBPQWALRECEIVER_CONNECT "Waiting in WAL receiver to establish connection to remote server." > LIBPQWALRECEIVER_RECEIVE "Waiting in WAL receiver to receive data from remote server." > SSL_OPEN_SERVER "Waiting for SSL while attempting connection." > WAIT_FOR_STANDBY_CONFIRMATION "Waiting for WAL to be received and flushed by the physical standby." > -WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary." > +WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary or standby." > WAIT_FOR_WAL_REPLAY "Waiting for WAL replay to reach a target LSN on a standby." > +WAIT_FOR_WAL_WRITE "Waiting for WAL write to reach a target LSN on a standby." > WAL_SENDER_WAIT_FOR_WAL "Waiting for WAL to be flushed in WAL sender process." > WAL_SENDER_WRITE_DATA "Waiting for any activity when processing replies from WAL receiver in WAL sender process." > > ABI_compatibility: WaitEventClient is about waiting for a socket to become readable or writable, so I think WAIT_FOR_WAL_* events don't fit in its scope. Sockets are just one of the ways to be in WAIT_FOR_WAL_*; other delay sources include local fsync and local replay, which could be disk-bound or CPU-bound. I think WAIT_FOR_WAL_* belong in WaitEventIPC. In the absence of objections, I'll change it that way: --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -91,5 +91,2 @@ SSL_OPEN_SERVER "Waiting for SSL while attempting connection." WAIT_FOR_STANDBY_CONFIRMATION "Waiting for WAL to be received and flushed by the physical standby." -WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary or standby." -WAIT_FOR_WAL_REPLAY "Waiting for WAL replay to reach a target LSN on a standby." -WAIT_FOR_WAL_WRITE "Waiting for WAL write to reach a target LSN on a standby." WAL_SENDER_WAIT_FOR_WAL "Waiting for WAL to be flushed in WAL sender process." @@ -164,2 +161,5 @@ SAFE_SNAPSHOT "Waiting to obtain a valid snapshot for a READ ONLY DEFER SYNC_REP "Waiting for confirmation from a remote server during synchronous replication." +WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary or standby." +WAIT_FOR_WAL_REPLAY "Waiting for WAL replay to reach a target LSN on a standby." +WAIT_FOR_WAL_WRITE "Waiting for WAL write to reach a target LSN on a standby." WAL_RECEIVER_EXIT "Waiting for the WAL receiver to exit." As further rationale, WaitEventClient says "never used for a background process". By xlogwait.c sizing its shmem to include NUM_AUXILIARY_PROCS, it's reserving the right to accept calls from background processes. As a side note related to that, the placement of WaitLSNCleanup() in ProcKill() but not AuxiliaryProcKill() would also need to change before welcoming auxiliary process use. Perhaps better than adding to AuxiliaryProcKill(), a separate on_shmem_exit callback for xlogwait.c would make it harder to miss callers needing it.
Re: wait_event_type for WAIT FOR LSN
От:
Xuneng Zhou <xunengzhou@gmail.com>
Дата:
Hi Noah, On Mon, Jul 6, 2026 at 9:26 AM Noah Misch wrote: > > commit 7a39f43 wrote: > > --- a/src/backend/utils/activity/wait_event_names.txt > > +++ b/src/backend/utils/activity/wait_event_names.txt > > @@ -76,34 +76,35 @@ ABI_compatibility: > > # Wait Events - Client > > # > > # Use this category when a process is waiting to send data to or receive data > > # from the frontend process to which it is connected. This is never used for > > # a background process, which has no client connection. > > # > > > > Section: ClassName - WaitEventClient > > > > CLIENT_READ "Waiting to read data from the client." > > CLIENT_WRITE "Waiting to write data to the client." > > GSS_OPEN_SERVER "Waiting to read data from the client while establishing a GSSAPI session." > > LIBPQWALRECEIVER_CONNECT "Waiting in WAL receiver to establish connection to remote server." > > LIBPQWALRECEIVER_RECEIVE "Waiting in WAL receiver to receive data from remote server." > > SSL_OPEN_SERVER "Waiting for SSL while attempting connection." > > WAIT_FOR_STANDBY_CONFIRMATION "Waiting for WAL to be received and flushed by the physical standby." > > -WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary." > > +WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary or standby." > > WAIT_FOR_WAL_REPLAY "Waiting for WAL replay to reach a target LSN on a standby." > > +WAIT_FOR_WAL_WRITE "Waiting for WAL write to reach a target LSN on a standby." > > WAL_SENDER_WAIT_FOR_WAL "Waiting for WAL to be flushed in WAL sender process." > > WAL_SENDER_WRITE_DATA "Waiting for any activity when processing replies from WAL receiver in WAL sender process." > > > > ABI_compatibility: > > WaitEventClient is about waiting for a socket to become readable or writable, > so I think WAIT_FOR_WAL_* events don't fit in its scope. Sockets are just one > of the ways to be in WAIT_FOR_WAL_*; other delay sources include local fsync > and local replay, which could be disk-bound or CPU-bound. > > I think WAIT_FOR_WAL_* belong in WaitEventIPC. In the absence of objections, > I'll change it that way: Thanks for raising this. It makes sense to me to move these events to the WaitEventIPC session. I'm wondering whether this reasoning also applies to WAIT_FOR_STANDBY_CONFIRMATION. > --- a/src/backend/utils/activity/wait_event_names.txt > +++ b/src/backend/utils/activity/wait_event_names.txt > @@ -91,5 +91,2 @@ SSL_OPEN_SERVER "Waiting for SSL while attempting connection." > WAIT_FOR_STANDBY_CONFIRMATION "Waiting for WAL to be received and flushed by the physical standby." > -WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary or standby." > -WAIT_FOR_WAL_REPLAY "Waiting for WAL replay to reach a target LSN on a standby." > -WAIT_FOR_WAL_WRITE "Waiting for WAL write to reach a target LSN on a standby." > WAL_SENDER_WAIT_FOR_WAL "Waiting for WAL to be flushed in WAL sender process." > @@ -164,2 +161,5 @@ SAFE_SNAPSHOT "Waiting to obtain a valid snapshot for a READ ONLY DEFER > SYNC_REP "Waiting for confirmation from a remote server during synchronous replication." > +WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary or standby." > +WAIT_FOR_WAL_REPLAY "Waiting for WAL replay to reach a target LSN on a standby." > +WAIT_FOR_WAL_WRITE "Waiting for WAL write to reach a target LSN on a standby." > WAL_RECEIVER_EXIT "Waiting for the WAL receiver to exit." > > > As further rationale, WaitEventClient says "never used for a background > process". By xlogwait.c sizing its shmem to include NUM_AUXILIARY_PROCS, it's > reserving the right to accept calls from background processes. As a side note > related to that, the placement of WaitLSNCleanup() in ProcKill() but not > AuxiliaryProcKill() would also need to change before welcoming auxiliary > process use. Perhaps better than adding to AuxiliaryProcKill(), a separate > on_shmem_exit callback for xlogwait.c would make it harder to miss callers > needing it. Good catch! Currently, we have no actual background processes to make use of the wait-for facility, but things could change in the upcoming releases. A dedicated on_shmem_exit callback for xlogwait.c also makes sense to me. I'll prepare a patch for that. -- Regards, Xuneng Zhou HighGo Software Co., Ltd.
Re: wait_event_type for WAIT FOR LSN
От:
Xuneng Zhou <xunengzhou@gmail.com>
Дата:
On Tue, Jul 7, 2026 at 4:29 AM Noah Misch wrote: > > On Mon, Jul 06, 2026 at 01:41:35PM +0800, Xuneng Zhou wrote: > > On Mon, Jul 6, 2026 at 9:26 AM Noah Misch wrote: > > > commit 7a39f43 wrote: > > > > --- a/src/backend/utils/activity/wait_event_names.txt > > > > +++ b/src/backend/utils/activity/wait_event_names.txt > > > > @@ -76,34 +76,35 @@ ABI_compatibility: > > > > # Wait Events - Client > > > > # > > > > # Use this category when a process is waiting to send data to or receive data > > > > # from the frontend process to which it is connected. This is never used for > > > > # a background process, which has no client connection. > > > > # > > > > > > > > Section: ClassName - WaitEventClient > > > > > > > > CLIENT_READ "Waiting to read data from the client." > > > > CLIENT_WRITE "Waiting to write data to the client." > > > > GSS_OPEN_SERVER "Waiting to read data from the client while establishing a GSSAPI session." > > > > LIBPQWALRECEIVER_CONNECT "Waiting in WAL receiver to establish connection to remote server." > > > > LIBPQWALRECEIVER_RECEIVE "Waiting in WAL receiver to receive data from remote server." > > > > SSL_OPEN_SERVER "Waiting for SSL while attempting connection." > > > > WAIT_FOR_STANDBY_CONFIRMATION "Waiting for WAL to be received and flushed by the physical standby." > > > > -WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary." > > > > +WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary or standby." > > > > WAIT_FOR_WAL_REPLAY "Waiting for WAL replay to reach a target LSN on a standby." > > > > +WAIT_FOR_WAL_WRITE "Waiting for WAL write to reach a target LSN on a standby." > > > > WAL_SENDER_WAIT_FOR_WAL "Waiting for WAL to be flushed in WAL sender process." > > > > WAL_SENDER_WRITE_DATA "Waiting for any activity when processing replies from WAL receiver in WAL sender process." > > > > > > > > ABI_compatibility: > > > > > > WaitEventClient is about waiting for a socket to become readable or writable, > > > so I think WAIT_FOR_WAL_* events don't fit in its scope. Sockets are just one > > > of the ways to be in WAIT_FOR_WAL_*; other delay sources include local fsync > > > and local replay, which could be disk-bound or CPU-bound. > > > > > > I think WAIT_FOR_WAL_* belong in WaitEventIPC. In the absence of objections, > > > I'll change it that way: > > > > Thanks for raising this. It makes sense to me to move these events to > > the WaitEventIPC session. I'm wondering whether this reasoning also > > applies to WAIT_FOR_STANDBY_CONFIRMATION. > > It partially does. In favor of WAIT_FOR_STANDBY_CONFIRMATION in WaitEventIPC: > > - It's much like SYNC_REP, which is in WaitEventIPC. > - It's a condition variable wait, not a socket wait. > > In favor of keeping WAIT_FOR_STANDBY_CONFIRMATION in WaitEventClient: > > - The standard for reclassifying a wait event post-release is higher, since it > impedes comparing waits across versions. > - The condition variable is set based on a socket read, and it's fair to treat > the condition variable layer as an implementation detail of little concern > to users. > > Overall, I lean toward not changing WAIT_FOR_STANDBY_CONFIRMATION. Thanks for clarification. I'm also ok with the status quo. -- Regards, Xuneng Zhou HighGo Software Co., Ltd.
Re: wait_event_type for WAIT FOR LSN
От:
Alexander Korotkov <aekorotkov@gmail.com>
Дата:
Hi! On Tue, Jul 7, 2026 at 5:08 PM Xuneng Zhou wrote: > > On Tue, Jul 7, 2026 at 4:29 AM Noah Misch wrote: > > > > On Mon, Jul 06, 2026 at 01:41:35PM +0800, Xuneng Zhou wrote: > > > On Mon, Jul 6, 2026 at 9:26 AM Noah Misch wrote: > > > > commit 7a39f43 wrote: > > > > > --- a/src/backend/utils/activity/wait_event_names.txt > > > > > +++ b/src/backend/utils/activity/wait_event_names.txt > > > > > @@ -76,34 +76,35 @@ ABI_compatibility: > > > > > # Wait Events - Client > > > > > # > > > > > # Use this category when a process is waiting to send data to or receive data > > > > > # from the frontend process to which it is connected. This is never used for > > > > > # a background process, which has no client connection. > > > > > # > > > > > > > > > > Section: ClassName - WaitEventClient > > > > > > > > > > CLIENT_READ "Waiting to read data from the client." > > > > > CLIENT_WRITE "Waiting to write data to the client." > > > > > GSS_OPEN_SERVER "Waiting to read data from the client while establishing a GSSAPI session." > > > > > LIBPQWALRECEIVER_CONNECT "Waiting in WAL receiver to establish connection to remote server." > > > > > LIBPQWALRECEIVER_RECEIVE "Waiting in WAL receiver to receive data from remote server." > > > > > SSL_OPEN_SERVER "Waiting for SSL while attempting connection." > > > > > WAIT_FOR_STANDBY_CONFIRMATION "Waiting for WAL to be received and flushed by the physical standby." > > > > > -WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary." > > > > > +WAIT_FOR_WAL_FLUSH "Waiting for WAL flush to reach a target LSN on a primary or standby." > > > > > WAIT_FOR_WAL_REPLAY "Waiting for WAL replay to reach a target LSN on a standby." > > > > > +WAIT_FOR_WAL_WRITE "Waiting for WAL write to reach a target LSN on a standby." > > > > > WAL_SENDER_WAIT_FOR_WAL "Waiting for WAL to be flushed in WAL sender process." > > > > > WAL_SENDER_WRITE_DATA "Waiting for any activity when processing replies from WAL receiver in WAL sender process." > > > > > > > > > > ABI_compatibility: > > > > > > > > WaitEventClient is about waiting for a socket to become readable or writable, > > > > so I think WAIT_FOR_WAL_* events don't fit in its scope. Sockets are just one > > > > of the ways to be in WAIT_FOR_WAL_*; other delay sources include local fsync > > > > and local replay, which could be disk-bound or CPU-bound. > > > > > > > > I think WAIT_FOR_WAL_* belong in WaitEventIPC. In the absence of objections, > > > > I'll change it that way: > > > > > > Thanks for raising this. It makes sense to me to move these events to > > > the WaitEventIPC session. I'm wondering whether this reasoning also > > > applies to WAIT_FOR_STANDBY_CONFIRMATION. > > > > It partially does. In favor of WAIT_FOR_STANDBY_CONFIRMATION in WaitEventIPC: > > > > - It's much like SYNC_REP, which is in WaitEventIPC. > > - It's a condition variable wait, not a socket wait. > > > > In favor of keeping WAIT_FOR_STANDBY_CONFIRMATION in WaitEventClient: > > > > - The standard for reclassifying a wait event post-release is higher, since it > > impedes comparing waits across versions. > > - The condition variable is set based on a socket read, and it's fair to treat > > the condition variable layer as an implementation detail of little concern > > to users. > > > > Overall, I lean toward not changing WAIT_FOR_STANDBY_CONFIRMATION. > > Thanks for clarification. I'm also ok with the status quo. Thanks to Noah for raising this. Thanks to Xuneng for the feedback. OK, let's leave WAIT_FOR_STANDBY_CONFIRMATION as is, but move WAIT_FOR_WAL_* to WaitEventIPC. The patch is attached. I'm going to push it if no objections. ------ Regards, Alexander Korotkov Supabase
