Horton

RE: Synchronizing slots from primary to standby

Hayato Kuroda (Fujitsu) <kuroda.hayato@fujitsu.com>

View thread View on postgr.es
From: "Hayato Kuroda (Fujitsu)" <kuroda.hayato@fujitsu.com>
To: 'shveta malik' <shveta.malik@gmail.com>, "Zhijie Hou (Fujitsu)" <houzj.fnst@fujitsu.com>, Amit Kapila <amit.kapila16@gmail.com>
Cc: Peter Smith <smithpb2250@gmail.com>, "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com>, Nisha Moond <nisha.moond412@gmail.com>, Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>, Peter Eisentraut <peter.eisentraut@enterprisedb.com>, Bruce Momjian <bruce@momjian.us>, Ashutosh Sharma <ashu.coek88@gmail.com>, Andres Freund <andres@anarazel.de>, pgsql-hackers <pgsql-hackers@postgresql.org>, Ajin Cherian <itsajin@gmail.com>, Alvaro Herrera <alvherre@alvh.no-ip.org>
Date: 2023-12-21T13:07:46Z
Lists: pgsql-hackers 
Dear Shveta,

Thanks for updating the patch! Here is my comments for v52-0002.

~~~~~
system-views.sgml

01. 

```
+
+     <row>
+      <entry role="catalog_table_entry"><para role="column_definition">
+       <structfield>sync_state</structfield> <type>char</type>
+      </para>
+      <para>
+      Defines slot synchronization state. This is meaningful on the physical
+      standby which has configured <xref linkend="guc-enable-syncslot"/> = true.
+      Possible values are:
+       <itemizedlist>
+        <listitem>
+         <para><literal>n</literal> = none for user created slots,
...
```

Hmm. I'm not sure why we must show a single character to a user. I'm OK for
pg_subscription.srsubstate because it is a "catalog" - the actual value would be
recorded in the heap. But pg_replication_slot is just a view so that we can replace
internal representations to other strings. E.g., pg_replication_slots.wal_status.
How about using {none, initialized, ready} or something?

~~~~~
postmaster.c

02. bgworker_should_start_now

```
+            if (start_time == BgWorkerStart_ConsistentState_HotStandby &&
+                     pmState != PM_RUN)
+                return true;
```

I'm not sure the second condition is really needed. The line will be executed when
pmState is PM_HOT_STANDBY. Is there a possibility that pmState is changed around here?

~~~~~
libpqwalreceiver.c

03. PQWalReceiverFunctions

```
+    .walrcv_get_dbname_from_conninfo = libpqrcv_get_dbname_from_conninfo,
```

Just to confirm - is there a rule for ordering?

~~~~~
slotsync.c

04. SlotSyncWorkerCtx

```
typedef struct SlotSyncWorkerCtx
{
	pid_t		pid;
	slock_t		mutex;
} SlotSyncWorkerCtx;

SlotSyncWorkerCtx *SlotSyncWorker = NULL;
```

Per other files like launcher.c, should we use a name like "SlotSyncWorkerCtxStruct"?

05. SlotSyncWorkerRegister()

Your coding will work well, but there is another approach which validates
slotsync parameters here. In this case, the postmaster should exit ASAP. This can
notify that there are some wrong settings to users earlier. Thought?

06. wait_for_primary_slot_catchup

```
+        CHECK_FOR_INTERRUPTS();
+
+        /* Handle any termination request if any */
+        ProcessSlotSyncInterrupts(wrconn);
```

ProcessSlotSyncInterrupts() also has CHECK_FOR_INTERRUPTS(), so no need to call.

07. wait_for_primary_slot_catchup

```
+        /*
+         * XXX: Is waiting for 2 seconds before retrying enough or more or
+         * less?
+         */
+        rc = WaitLatch(MyLatch,
+                       WL_LATCH_SET | WL_TIMEOUT | WL_POSTMASTER_DEATH,
+                       2000L,
+                       WAIT_EVENT_REPL_SLOTSYNC_PRIMARY_CATCHUP);
+
+        ResetLatch(MyLatch);
+
+        /* Emergency bailout if postmaster has died */
+        if (rc & WL_POSTMASTER_DEATH)
+            proc_exit(1);
```

Is there any reasons not to use WL_EXIT_ON_PM_DEATH event? If not, you can use.

08. synchronize_slots

```
+    SpinLockAcquire(&WalRcv->mutex);
+    if (!WalRcv ||
+        (WalRcv->slotname[0] == '\0') ||
+        XLogRecPtrIsInvalid(WalRcv->latestWalEnd))
+    {
...
```

Assuming that WalRcv is still NULL. In this case, does the first SpinLockAcquire()
lead a segmentation fault?

09. synchronize_slots

```
+    elog(DEBUG2, "slot sync worker's query:%s \n", s.data);
```

The query is not dynamical one, so I think no need to print even if the debug
mode.

10. synchronize_one_slot

IIUC, this function can synchronize slots even if the used plugin on primary is
not installed on the secondary server. If the slot is created by the slotsync
worker, users will recognize it after the server is promoted and the decode is
starting. I felt it is not good specification. Can we detect in the validation
phase?

~~~~~
not the source code

11. 

I tested the typical case - promoting a publisher from a below diagram.
A physical replication slot "physical" was specified as standby_slot_names.

```
node A (primary) --> node B (secondary)
|
|
node C (subscriber)
```

And after the promoting, below lines were periodically output on logfiles for
node B and C.

```
WARNING:  replication slot "physical" specified in parameter "standby_slot_names" does not exist, ignoring
```

Do you have idea to suppress the warning? IIUC it is a normal behavior of the
walsender so that we cannot avoid the periodical outputs.

The steps of the test was as follows:

1. stop the node A via pg_ctl stop
2. promota the node B via pg_ctl promote
3. change the connection string of the subscription via ALTER SUBSCRIPTION ... CONNECTION ...

Best Regards,
Hayato Kuroda
FUJITSU LIMITED

Commits

  1. Doc: Add the new section "Logical Replication Failover".

  2. Fix the review comments and a bug in the slot sync code.

  3. Fix a test in failover slots regression test.

  4. Fix the intermittent buildfarm failures in 040_standby_failover_slots_sync.

  5. Split XLogCtl->LogwrtResult into separate struct members

  6. Ensure that the sync slots reach a consistent state after promotion without losing data.

  7. Introduce a new GUC 'standby_slot_names'.

  8. Fix BF failure introduced by commit b3f6b14cf4.

  9. Fixups for commit 93db6cbda0.

  10. Fix BF failure in commit 93db6cbda0.

  11. Add a new slot sync worker to synchronize logical slots.

  12. Improve ERROR/LOG messages added by commits ddd5f4f54a and 7a424ece48.

  13. Disable autovacuum on primary in 040_standby_failover_slots_sync test.

  14. Fix the incorrect format specifier used in commit 7a424ece48.

  15. Change the LOG level in 040_standby_failover_slots_sync.pl to DEBUG2.

  16. Add more LOG and DEBUG messages for slot synchronization.

  17. Another try to fix BF failure introduced in commit ddd5f4f54a.

  18. Fix BF introduced in commit ddd5f4f54a.

  19. Add a slot synchronization function.

  20. Clean-ups for 776621a5e4 and 7329240437.

  21. Enhance libpqrcv APIs to support slot synchronization.

  22. Add a failover option to subscriptions.

  23. Allow setting failover property in the replication command.

  24. Allow to enable failover property for replication slots via SQL API.

  25. Add macros for looping through a List without a ListCell.

  26. Track conflict_reason in pg_replication_slots.

  27. Allow upgrades to preserve the full subscription's state.

  28. Add support for incremental backup.

  29. Simplify some logic in CreateReplicationSlot()

  30. Fix uninitialized access to InitialRunningXacts during decoding after ERROR.

  31. Flexible options for CREATE_REPLICATION_SLOT.

  32. Empty search_path in logical replication apply worker and walsender.

  33. Fix oldest xmin and LSN computation across repslots after advancing

  34. Require C99 (and thus MSCV 2013 upwards).