Re: when set track_commit_timestamp on, database system abort startup

On Wed, Sep 19, 2018 at 12:12:44PM +0900, Kyotaro HORIGUCHI wrote:
> The fix looks good to me. The TAP test works fine.
>
> In the TAP test:
>
> ====
> The test script lacks a general description about its objective.

That's always welcome. The patch looks sensible to me.

> The following error message is seen when the issue happens.
>
>> DETAIL: Could not read from file "pg_commit_ts/0000" at offset 8192: Success.
>
> This seems what 811b6e36a9 tried to fix. This should be like the follows.
>
>> DETAIL: Could not read from file "pg_commit_ts/0000" at offset 8192: read 0 of 8192"
>
> I, as one of reviewers of it, didn't remember how it was
> overlooked butthis also needs a fix in this patch or as a
> separate issue.

Not completely overlooked... Please see here why I have left out the
bits in slru.c, on purpose:
https://www.postgresql.org/message-id/20180613035038.GE3138@paquier.xyz
If you want to address that as well, please feel free to send a patch.
Reporting the number of bytes read would be the nicest error report, but
this requires a new field similar to slru_errno which includes an error
string to use if errno is 0.
--
Michael