Starting a few days ago we started to see a strange behavior where writing to the database causes massive read operations. For example, I have a table that needs to be updated every night (about 20,000 rows). Using Django ORM, we update them one item at a time. Usually the overall process takes a few minutes; it is now taking hours (like over 15 hours).
Running atop, we're seeing Read Disk values in the range of 147.2M/10s and Write Disk values in the range of 16K/10s. Together, the Disk throughput is in the high 90% and frequently hits 100%.
When I'm not writing to the database, it becomes quiet without any disk activity.
This is a 256 MByte system and atop shows we're not swapping
Is autovacuum set to 'on'?
Possibly there is bloat in your table and indexes; what happens to the system after you issue a manual "VACUUM ANALYZE VERBOSE"? If that helps, you might need to make your autovacuum settings more aggressive.
