Do not retry errno 2013 in tablet #9009
Conversation
these non-retryable in the tablet. Other cases that lead to 2013 like auth and packet size limits are unlikely to succeed on retry anyway. One exception might be network errors, but we should we be papering over these, anyway? Signed-off-by: Jacques Grove <aquarapid@gmail.com>
Signed-off-by: Jacques Grove <aquarapid@gmail.com>
aquarapid
added
Type: Enhancement
Signed-off-by: Jacques Grove <aquarapid@gmail.com>
There was a problem hiding this comment.
Looks reasonable to me.
@harshit-gangal can you take a look? Should we be concerned about backwards-compatibility?
|
I remember we use this to re-try traffic for reserved connection cases if they gets killed due to inactivity on connection. |
|
This is on vttablet side, go ahead with the changes. |
To be clear, you are talking about vtgate here, whereas the PR is for vttablet. Is that correct? |
Yes, not a problem here. |
Do not retry "server lost" (errno 2013) errors in vttablet.
These errors are usually due to:
Usually not helpful to retry the above.
One exception are certain network issues (have to be serious to break a TCP conn), that we may paper over with current retries. I would argue that this is not a great idea; or at least we should be more noisy about it (right now these retries only show up in metrics; users are very unlikely to notice).
One option could be to instead make this configurable, to preserve existing behavior by default. Comments welcome.