Expose host address when timeout exceeded #1943
Conversation
Fixes: jackc#1942 Once we face a timeout we should expose address, it can be for example ipv6 or ipv4 and knowing what exact address we took for resolver extreamly helpful for triaging a networking issue. This issue is also fixed in pgx/v4, see: jackc/pgconn#139
|
I don't think this is right. errTimeout can some sort of TCP timeout. It is entirely possible that one connection attempt to one host will timeout without the entire connection attempt to all hosts having timed out due to context cancellation or some other global timeout. |
|
It's not possible with current sequential logic, once we receive timeout all other fallback will receive same timeout error |
pgconn/pgconn.go
Outdated
| if _, ok := cerr.err.(*errTimeout); ok { | ||
| // once we reach timeout it's useless to check other fallbacks | ||
| break | ||
| } |
There was a problem hiding this comment.
What I'm saying is it's not always useless to check other fallbacks. errTimeout could come from a TCP timeout. If the context has not expired the next fallback should be tried.
There was a problem hiding this comment.
Got it.
I made one more attempt :D
So now I do a bit different thing, I keep track on fallback error and override result error only if needed.
For a deadline errors we should keep only first deadline error in result, every next one is false-positive errors (since we not actually do anything).
|
This one is addressed here: 8db9716 |
Fixes: #1942
Once we face a timeout we should expose address, it can be for example ipv6 or ipv4 and knowing what exact address we took for resolver extreamly helpful for triaging a networking issue.
This issue is also fixed in pgx/v4, see: jackc/pgconn#139