Re: recovery test failure on morepork with timestamp mystery

On 2022-05-13 04:14, Andres Freund wrote:

> One interesting tidbit is that the log timestamps are computed differently
> (with elog.c:get_formatted_log_time()) than the reset timestamp (with
> GetCurrentTimestamp()). Both use gettimeofday() internally.
> 
> I wonder if there's a chance that somehow openbsd ends up with more usecs than
> "fit" in a second in the result of gettimeofday()? The elog.c case would
> truncate everything above a second away afaics:
>     /* 'paste' milliseconds into place... */
>     sprintf(msbuf, ".%03d", (int) (saved_timeval.tv_usec / 1000));
>     memcpy(formatted_log_time + 19, msbuf, 4);
> 
> whereas GetCurrentTimestamp() would add them to the timestamp:
>     result = (TimestampTz) tp.tv_sec -
>         ((POSTGRES_EPOCH_JDATE - UNIX_EPOCH_JDATE) * SECS_PER_DAY);
>     result = (result * USECS_PER_SEC) + tp.tv_usec;
> 

Well, I don't know if you remember but there was a thread a while back 
and a test program (monotime.c) to test the clock if it could go 
backwards and openbsd showed the following result when running the 
attached testprogram:

openbsd 5.9:

$ ./monotime
1021006 Starting
1017367 Starting
1003415 Starting
1007598 Starting
1021006 Stopped
1007598 Stopped
1017367 Stopped
1003415 Stopped

openbsd 6.9:

$ ./monotime
410310 Starting
547727 Starting
410310 Back 262032.372314102 => 262032.242045208
410310 Stopped
465180 Starting
255646 Starting
547727 Stopped
465180 Stopped
255646 Stopped

could that have something to do with it?

/Mikael