Re: BUG #18241: PushTransaction may cause Standby to execute ItemIdMarkDead

Kyotaro Horiguchi <horikyota.ntt@gmail.com>

From: Kyotaro Horiguchi <horikyota.ntt@gmail.com>

To: feichanghong@qq.com

Cc: pgsql-bugs@lists.postgresql.org

Date: 2023-12-12T04:49:00Z

Lists: pgsql-bugs

Commits

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

Prevent tuples to be marked as dead in subtransactions on standbys
- 4d45ecc9284a 12.18 landed
- d99294e77f1d 13.14 landed
- 0cfd3ddfe070 14.11 landed
- f5d8f59cae3e 15.6 landed
- 0e2c05af90ee 16.2 landed
- 8a7cbfce13d4 17.0 landed
Allow read only connections during recovery, known as Hot Standby.
- efc16ea52067 9.0.0 cited

At Mon, 11 Dec 2023 20:13:23 +0800, "feichanghong" <feichanghong@qq.com> wrote in 
> &gt; We have analyzed the cause of the problem: The value of
> &gt;&nbsp;IndexScanDesc-&gt;xactStartedInRecovery is a key condition for determining
> &gt;&nbsp;whether an index item can be marked as DEAD. And it depends on
> &gt;&nbsp;CurrentTransactionState-&gt;startedInRecovery. However, PushTransaction does
> &gt;&nbsp;not assign a value to startedInRecovery when modifying
> &gt;&nbsp;CurrentTransactionState.
> The attached patch has been verified to resolve the mentioned issue.

This appears to be a bug that has existed for a long time since commit
efc16ea520 (in 2009). Your fix looks correct to me, but as for me, the
comment is not particularly necessary, and it would be sufficient to
insert the new line in the location according to the member order
within TransactionStateData.

I have briefly checked and found no issues with other struct members
of TransactionStateData.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center