Обсуждение: WAL size many times data size?
Our streaming replication and WAL archiving keep having trouble because of the sheer size of the WAL files. The number of bytes in the WAL files seems to be a large multiplier, like 5x or 10x, the amount of data that we load. It's hard to know exactly, because in addition to the size of the actual data files, there are indexes and auxiliary tables of crunched data. But these shouldn't even double the total data.
The other day we loaded about 5-8 GB of data, and the WAL directory ended up with 75GB in it. Streaming replication to our standby server broke because the network couldn't keep up, so it had to fall back to the secondary stream of WAL files.
Is this expected?
Thanks,
Craig
Craig James <cjames@emolecules.com> writes:
> Our streaming replication and WAL archiving keep having trouble because of
> the sheer size of the WAL files. The number of bytes in the WAL files seems
> to be a large multiplier, like 5x or 10x, the amount of data that we load.
My first instinct would be to check how often checkpoints are happening,
and increase the checkpoint control parameters so that they're at least
5 to 15 minutes apart. A short checkpoint interval leads directly to
WAL bloat because it means more full-page images get inserted into WAL.
(The first change of any given page after a checkpoint requires an FPI.)
regards, tom lane