hyperledger-labs / fabric-smart-client / 40498463864

91%

Build Type

push

github

Committed by web-flow

Commit Message

fix(storage): map a Postgres connect timeout to a lost connection (#2277)

A Postgres failover longer than committer.retries × committer.retrySleep stopped a channel's block stream when connect_timeout was set in the dataSource. pgx reports an expired connect_timeout as a *pgconn.ConnectError wrapping context.DeadlineExceeded. ErrorMapper.WrapError returned every error matching context.DeadlineExceeded unchanged, so the committer counted the attempt against its retry budget instead of retrying it as a lost connection.

WrapError now maps a failed connect to driver.ErrConnectionLost even when it ended with context.DeadlineExceeded. The connect_timeout context derives from the caller's, so pgx reports a caller deadline that expires while connecting with the same error. A caller that owns a deadline therefore checks its own context before retrying, as retryBlock does. A cancelled context stays unchanged, so shutdown is not retried. A SQLSTATE the server reports while connecting is looked up before the connect error, so a rejected password or a missing database stays unmapped and the channel degrades instead of retrying a misconfiguration forever.

TestErrorMapper_ConnectTimeout connects through database/sql to a listener that never answers. It covers connect_timeout, a two-host DSN where one host times out and the other refuses, a caller deadline, a caller cancel, and a fake server that rejects the password with 28P01.

A 53300 (too_many_connections) on connect is still returned unchanged; #2278 tracks it.

Fixes #2270

Signed-off-by: Marcus Brandenburger <bur@zurich.ibm.com>

Coverage Stats

17 of 17 new or added lines in 1 file covered. (100.0%)

13 existing lines in 6 files now uncovered.

30095 of 32930 relevant lines covered (91.39%)

1248.19 hits per line