| From: | cca5507 <cca5507(at)qq(dot)com> |
|---|---|
| To: | Bertrand Drouvot <bertranddrouvot(dot)pg(at)gmail(dot)com> |
| Cc: | Michael Paquier <michael(at)paquier(dot)xyz>, pgsql-hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org> |
| Subject: | Re: Historic snapshot doesn't track txns committed in BUILDING_SNAPSHOT state |
| Date: | 2024-08-12 11:38:53 |
| Message-ID: | tencent_A6221937BC93033D3B89D096C2A45A839407@qq.com |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
Hi,
Thanks for the comments!
- I think it's fine to skip during fast forward as we are not generating logical
- changes. It's done that way in master, in your proposal and in my "if" proposals.
- Note that my proposals related to the if conditions are for heap2_decode and
- heap_decode (not xact_decode).
But note that in xact_decode(), case XLOG_XACT_INVALIDATIONS, we call
ReorderBufferXidSetCatalogChanges() even if we are fast-forwarding, it might
be better to be consistent.
In addition, we don't decode anything during fast forward, but the snapshot might
serialize to disk. If we skip calling ReorderBufferXidSetCatalogChanges(), the snapshot
may be wrong on disk.
--
Regards,
ChangAo Chen
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Amit Kapila | 2024-08-12 11:40:47 | Re: Conflict detection and logging in logical replication |
| Previous Message | Aleksander Alekseev | 2024-08-12 11:32:42 | Re: format_datum debugging function |