RE: Slow catchup of 2PC (twophase) transactions on replica in LR

On Tuesday, July 16, 2024 1:17 PM Kuroda, Hayato/黒田 隼人 <kuroda.hayato@fujitsu.com> wrote
> 
> Dear Amit, Hou,
> 
> Thanks for giving comments! PSA new versions.
> What's new:
> 
> 0001: included Hou's patch [1] not to overwrite slot options.
>       Some other comments were also addressed.

Thanks for the patch!

One more issue I found is that:

+IsTwoPhaseTransactionGidForSubid(Oid subid, char *gid)
+{
+    int            ret;
+    Oid            subid_written;
+    TransactionId xid;
+
+    ret = sscanf(gid, "pg_gid_%u_%u", &subid_written, &xid);
+
+    return (ret == 2 && subid == subid_written);

I think it's not correct to use sscanf here, because it will return the same value
even if the gid is "pg_gid_123_123_123_123..." which isn't a
gid created by the apply worker. I think we should use TwoPhaseTransactionGid
to build the gid string and compare it with each existing gid(strcmp).

Best Regards,
Hou zj