Closed osrf-migration closed 7 years ago
Original comment by Hugo Boyer (Bitbucket: hugomatic, GitHub: hugomatic).
After a quick investigation with @chapulina, we decided to instrument the websocket events.
Since the sasc round is creating a lot of new resources on the portal (3 ssh keys, multiple machines), and each can trigger multiple updates. It's also interesting that a lot of those calls are 304 redirects, and I will try to understand how these work.
The logging output is to be customize to better understand the delays between the calls.
Original report (archived issue) by Hugo Boyer (Bitbucket: hugomatic, GitHub: hugomatic).
The following info shows the result of the node log
tail -f /var/log/nodejs/nodejs.log
that result from user interaction on the sasc kiosk.The log format could be better (there's a response time but no date, I will be fixing that shortly). In general, these calls are often happening one after the other.
user action: CREATE ROUND
user action: LAUNCH ARBITER
user action LAUNCH ALL BLUE PAYLOADS (1 machine):
user action: FINISH ROUND
If you eliminate a few refreshes (launching->running, find the ip), there are still a lot of unnecessary calls. Also, these calls used to be very slow (when the database was on a t2.micro).
Because we are using REST, PUT calls contain the whole state for each update, and the database incurs many writes.
The problem may be due to:
Solutions may include: