Re: long-standing data loss bug in initial sync of logical replication

On Thu, Mar 13, 2025 at 2:12 PM Hayato Kuroda (Fujitsu)
<kuroda(dot)hayato(at)fujitsu(dot)com> wrote:
>
>
> Workload C. DDL is happening on publication but on unrelated table
> ============================================
> We did not run the workload because we expected this could be same results as D.
> 588acf6 is needed to optimize the workload.
>
> -----
>
> Workload D. DDL is happening on the related published table,
> and one insert is done per invalidation
> =========================================
> This workload had huge regression same as the master branch. This is expected
> because distributed invalidation messages require all concurrent transactions
> to rebuild relsync caches.
>
> Concurrent txn | Head (sec) | Patch (sec) | Degradation (%)
> ------------------ | ------------ | ------------ | ----------------
> 50 | 0.013496 | 0.015588 | 15.5034
> 100 | 0.015112 | 0.018868 | 24.8517
> 500 | 0.018483 | 0.038714 | 109.4536
> 1000 | 0.023402 | 0.063735 | 172.3524
> 2000 | 0.031596 | 0.110860 | 250.8720
>

IIUC, workloads C and D will have regression in back branches, and
HEAD will have regression only for workload D. We have avoided
workload C regression in HEAD via commits 7c99dc587a and 3abe9dc188.
We can backpatch those commits if required, but I think it is better
not to do those as scenarios C and D won't be that common, and we
should go ahead with the fix as it is. In the future, if we get any
way to avoid regression due to scenario-D, then we can do that for the
HEAD branch.

Thoughts?

--
With Regards,
Amit Kapila.