Ascii graphs from gnuplot are great. Next challenge is better tuning for Postgres to reduce insert rate variance. Workload is insert only so vacuum should not be an issue. Config is github.com/mdcallag/mytoo
Next steps are:
* check Postgres log
* modify bgwriter_* options
Conversation
Messages from the log that suggest increasing WAL, but WAL is already 32g for a ~250g database. At what point would I be spending too much on storage for the WAL?
gist.github.com/mdcallag/ed0d7
I will repeat a test with full_page_write=off to see what changes, but I prefer =on.
3
Replying to
This happens because you set checkpoint_warning to 1800s. Most people don't increase it from the default setting of 30s (30 seconds).
Replying to
My question is whether checkpoints happening this frequently is something I really need to fix. While I appreciate feedback from PG experts like you, struggling a bit with it is also a good way for me to learn more about PG.
1
Replying to
Makes sense! Anyway, checkpoints that occur every 5+ minutes are typically not a problem.
1