Performance degradation in hot standby in HA
@wlodek recently implemented extended HA stability tests and the results are worrisome. It seems that there is a huge performance degradation observed in the following scenarios:
- hotstandby memfile v4 (after 40s)
- hotstandby memfile v6 (after 10s)
- hotstandby postgresql v4 (after 46s)
- hotstandby postgresql v6 (erratic, degradation over a period between 55 and 65s, but also slow degradation from 6000 acks/s down to 5500s in the initial 55s)
- loadbalancing postgresql v4 (43s)
Open the report report-ha-1.9.11-master.html, and go to HA stability tab. For example, see "ha multi hotstandby memfile v4". These are charts 1-A and 1-B. Note the violet line (ACK-received), which behaves correctly for roughly 40 seconds (~11k ACKs/sec) and then degrades rapidly to 3k ACKs/sec.
@wlodek will provide more details, including configs, once he gets back to office.
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information