On 08/11/2018 04:18 PM, Tomas Vondra wrote:
> On 08/11/2018 04:15 PM, Tom Lane wrote:
>> Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
>>>>> On 08/09/2018 07:47 PM, Alvaro Herrera wrote:
>>>>>> Actually, it seems to me that ApplyLogicalMappingFile is just leaking
>>>>>> the file descriptor for no good reason.
>>
>>> I think the fix can be as simple as attached ... I'm mostly afk for the
>>> weekend, so I'll commit & backpatch on Monday or so.
>>
>> LGTM. While you're at it, would you fix the misspelling three lines
>> below this?
>>
>> * Check whether the TransactionOId 'xid' is in the pre-sorted array 'xip'.
>> ^
>
> Sure.
>
I've pushed this, and backpatched it all the way back to 9.4 where
logical decoding was introduced.
While this resolves the way to run out of file descriptors, I wonder if
there are other trivial ways to trigger it (say, long-running
transaction spanning many 'vacuum full pg_class' runs). Not sure, will
try later.
The other question is whether errors are handled correctly - as reported
initially, I've seen this to trigger
ERROR: XX000: subtransaction logged without previous top-level txn record
I assume just fixing the error did not really fix that, so that if
something else fails we might end up in the same state.
regards
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services