Add retry support for serializable transactions #386
Conversation
|
This stack of pull requests is managed by Graphite. Learn more about stacking. |
neekolas
force-pushed
the
04-30-add_retry_support_for_serializable_transactions
branch
from
c65d2bc to
4c59544
Compare
| var err error | ||
| for i := 0; i < numRetries; i++ { | ||
| select { | ||
| case <-ctx.Done(): |
There was a problem hiding this comment.
noob Q: What does this select on this ctx do?
The default case returns nil when it finishes succesfully, but ctx seems like some sort of global context? When would it trigger?
There was a problem hiding this comment.
If you ctrl+c or the context is out of time, then ctx.Done() channel will receive a struct and this select block will return the error.
There was a problem hiding this comment.
Exactly. It just stops you from continuing retries when the server is in the process of shutting down
neekolas
deleted the
04-30-add_retry_support_for_serializable_transactions
branch
|
The specific tests that were failing are a little worrying to me because they all used random wallets and inboxes, and executed each test serially, so they should be isolated from one another. We may have to do some debugging around exactly what Postgres considers a "conflict" in the specific queries we are using here. Maybe it's a broader set of rows than you'd expect. But these retries are a generally good idea regardless and we can tackle that separately. |
tl;dr
I noticed our
libxmtptest suite was occasionally failing due to serializable transaction errors. This adds retries, which are required when using the serializable isolation level since errors are common and transient thanks to Postgres' conflict detection