Re: GIN improvements part 1: additional information - Mailing list pgsql-hackers

On 14.1.2014 00:38, Tomas Vondra wrote:
> On 13.1.2014 18:07, Alexander Korotkov wrote:
>> On Sat, Jan 11, 2014 at 6:15 AM, Tomas Vondra <tv@fuzzy.cz
>> <mailto:tv@fuzzy.cz>> wrote:
>>
>>     On 8.1.2014 22:58, Alexander Korotkov wrote:
>>     > Thanks for reporting. Fixed version is attached.
>>
>>     I've tried to rerun the 'archie' benchmark with the current patch, and
>>     once again I got
>>
>>        PANIC:  could not split GIN page, didn't fit
>>
>>     I reran it with '--enable-cassert' and with that I got
>>
>>     TRAP: FailedAssertion("!(ginCompareItemPointers(&items[i - 1],
>>                        &items[i]) < 0)", File: "gindatapage.c", Line: 149)
>>     LOG:  server process (PID 5364) was terminated by signal 6: Aborted
>>     DETAIL:  Failed process was running: INSERT INTO messages ...
>>
>>     so the assert in GinDataLeafPageGetUncompressed fails for some reason.
>>
>>     I can easily reproduce it, but my knowledge in this area is rather
>>     limited so I'm not entirely sure what to look for.
>>
>>
>> I've fixed this bug and many other bug. Now patch passes test suite that
>> I've used earlier. The results are so:
>
> OK, it seems the bug is gone. However now there's a memory leak
> somewhere. I'm loading pgsql mailing list archives (~600k messages)
> using this script
>
>    https://bitbucket.org/tvondra/archie/src/1bbeb920/bin/load.py
>
> And after loading about 1/5 of the data, all the memory gets filled by
> the pgsql backends (loading the data in parallel) and the DB gets killed
> by the OOM killer.

I've spent a fair amount of time trying to locate the memory leak, but
so far no luck. I'm not sufficiently familiar with the GIN code.

I can however demonstrate that it's there, and I have rather simple test
case to reproduce it - basically just a CREATE INDEX on a table with ~1M
email message bodies (in a tsvector column). The data is available here
(360MB compressed, 1GB raw):

   http://www.fuzzy.cz/tmp/message-b.data.gz

Simply create a single-column table, load data and create the index

   CREATE TABLE test ( body_tsvector TSVECTOR );
   COPY test FROM '/tmp/message-b.data';
   CREATE test_idx ON test USING gin test ( body_tsvector );

I'm running this on a machine with 8GB of RAM, with these settings

   shared_buffers=1GB
   maintenance_work_mem=1GB

According to top, CREATE INDEX from the current HEAD never consumes more
than ~25% of RAM:

    PID USER      PR  NI    VIRT    RES    SHR  %CPU %MEM  COMMAND
  32091 tomas     20   0 2026032 1,817g 1,040g  56,2 23,8  postgres

which is about right, as (shared_buffers + maintenance_work_mem) is
about 1/4 of RAM.

With the v5 patch version applied, the CREATE INDEX process eventually
goes crazy and allocates almost all the available memory (but somesimes
finishes, mostly by pure luck). This is what I was able to get from top

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM  COMMAND
  14090 tomas     20   0 7913820 6,962g 955036 D   4,3 91,1  postgres

while the system was still reasonably responsive.