Erik Jones <ejones@engineyard.com> writes:
> On Jan 26, 2015, at 2:43 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Maybe. That would require at least one crash recovery to have happened on
>> the server, which Erik seemed to be claiming had not happened. But
>> if there were any crashes then it would possibly fit.
> Definitely nothing recent crash/recovery-wise but digging further back in the logs it looks like there was a crash
recoveryrun startup back on 2014-09-30, which was about a week after this server was created. I suppose itâs
possiblethat the index has been corrupt since then but only now showing something visible?
Doesn't fit this specific bug fix --- the case it addresses would cause
already-existing rows to become unreachable from the index during crash
recovery. AFAICS that could not create a latent problem for rows inserted
later. Still, this isn't the only bug fixed in 9.3.4/9.3.5. Personally
I'm wondering about c0bd128c81c2b23a1cbc53305180fca51b3b61c3.
I'll try to refrain from asking why a server initdb'd in September wasn't
running 9.3.5 from inception.
regards, tom lane