Horton

Re: Minimal logical decoding on standbys

bdrouvotAWS <bdrouvot@amazon.com>

View thread
From: "Drouvot, Bertrand" <bdrouvot@amazon.com>
To: Alvaro Herrera <alvherre@2ndquadrant.com>, Andres Freund <andres@anarazel.de>
Cc: Ibrar Ahmed <ibrar.ahmad@gmail.com>, Amit Khandekar <amitdkhan.pg@gmail.com>, <fabriziomello@gmail.com>, tushar <tushar.ahuja@enterprisedb.com>, "[pgdg] Robert Haas" <robertmhaas@gmail.com>, Rahila Syed <rahila.syed@2ndquadrant.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2021-10-27T06:55:45Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Reduce the log level in 035_standby_logical_decoding.pl.

  2. 035_standby_logical_decoding: Add missing waits for replication

  3. For cascading replication, wake physical and logical walsenders separately

  4. Handle logical slot conflicts on standby

  5. Support invalidating replication slots due to horizon and wal_level

  6. Prevent use of invalidated logical slot in CreateDecodingContext()

  7. Replace replication slot's invalidated_at LSN with an enum

  8. Pass down table relation into more index relation functions

  9. Assert only valid flag bits are passed to visibilitymap_set()

  10. Remove unused _bt_delitems_delete() argument.

  11. Add xl_btree_delete optimization.

Attachments

Hi,

On 9/15/21 1:36 PM, Drouvot, Bertrand wrote:
>
> Hi,
>
> On 9/9/21 9:17 AM, Drouvot, Bertrand wrote:
>>
>> Hi Alvaro,
>>
>> On 8/2/21 4:56 PM, Drouvot, Bertrand wrote:
>>> Hi Alvaro,
>>>
>>> On 7/28/21 5:26 PM, Alvaro Herrera wrote:
>>>> On 2021-Jul-27, Drouvot, Bertrand wrote:
>>>>
>>>>> diff --git a/src/backend/utils/cache/lsyscache.c 
>>>>> b/src/backend/utils/cache/lsyscache.c
>>>>> +bool
>>>>> +get_rel_logical_catalog(Oid relid)
>>>>> +{
>>>>> +     bool    res;
>>>>> +     Relation rel;
>>>>> +
>>>>> +     /* assume previously locked */
>>>>> +     rel = table_open(relid, NoLock);
>>>>> +     res = RelationIsAccessibleInLogicalDecoding(rel);
>>>>> +     table_close(rel, NoLock);
>>>>> +
>>>>> +     return res;
>>>>> +}
>>>> So RelationIsAccessibleInLogicalDecoding() does a cheap check for
>>>> wal_level which can be done without opening the table; I think this
>>>> function should be rearranged to avoid doing that when not needed.
>>>
>>> Thanks for looking at it.
>>>
>>>
>>>> Also, putting this function in lsyscache.c seems somewhat wrong since
>>>> it's not merely accessing the system caches ...
>>>>
>>>> I think it would be better to move this elsewhere (relcache.c, 
>>>> proto in
>>>> relcache.h, perhaps call it 
>>>> RelationIdIsAccessibleInLogicalDecoding) and
>>>> short-circuit for the check that can be done before opening the table.
>>
>> So you have in mind to check for XLogLogicalInfoActive() first, and 
>> if true, then open the relation and call
>> RelationIsAccessibleInLogicalDecoding()?
>>
>> If so, then what about also creating a new 
>> RelationIsAccessibleWhileLogicalWalLevel() or something like this 
>> doing the same as RelationIsAccessibleInLogicalDecoding() but without 
>> the XLogLogicalInfoActive() check?
>>
>>>> At least the GiST code appears to be able to call this several 
>>>> times per
>>>> vacuum run, so it makes sense to short-circuit it for the fast case.
>>>>
>>>> ... though looking at the GiST code again I wonder if it would be more
>>>> sensible to just stash the table's Relation pointer somewhere in the
>>>> context structs 
>>
>> Do you have a "good place" in mind?
>>
> Another rebase attached.
>
> The patch proposal to address Andre's walsender corner cases is still 
> a dedicated commit (as i think it may be easier to discuss).
>
Another rebase attached (mainly to fix TAP tests failing due to b3b4d8e68a).

@Andres, the patch file number 6 contains an attempt to fix the 
Walsender corner case you pointed out.

@Alvaro, I did not look at your remark yet. Do you have a "good place" 
in mind? (related to "just stash the table's Relation pointer somewhere 
in the context structs")

Given the size of this patch series, I'm wondering if we could start 
committing piece per piece (while still working on the corner cases in 
parallel).

That would maximize the amount of coverage it gets in the v15 
development cycle.

What do you think?

Thanks

Bertrand