hyperledger-labs / fabric-smart-client / 40507528697

91%

Build Type

push

github

Committed by web-flow

Commit Message

fix(storage): retry a Postgres too_many_connections on connect (#2280)

A Postgres server at max_connections rejects a new connection with SQLSTATE 53300 (too_many_connections). That is common right after a failover, when every client reconnects to the new primary at once. ErrorMapper.WrapError had no entry for 53300 and returned it unchanged, so the committer degraded the block instead of retrying it.

WrapError now maps 53300 to driver.ErrConnectionLost, as it maps 57P03 (cannot_connect_now), and retryBlock retries it with backoff until the server accepts the connection or the node shuts down. SqlBusy would give it the bounded retry budget, which a reconnect storm can outlast. The other class 53 codes report a limit hit by a running statement. Reconnecting does not clear them, so they stay unmapped.

pgx joins the errors of every connect attempt into one *pgconn.ConnectError, with one or two attempts per host. WrapError maps a failed connect to driver.ErrConnectionLost when any attempt reported a SQLSTATE that maps to it, whichever host is listed first. Otherwise the first SQLSTATE decides, so a rejected password stays unmapped. An attempt without a SQLSTATE does not override another attempt's SQLSTATE: under sslmode=prefer, pgx records such an attempt for a TLS handshake the server refused before it retries the same host in plaintext.

TestErrorMapper_ConnectTimeout is renamed to TestErrorMapper_Connect and gains fake servers that refuse TLS and reject the startup message with a given SQLSTATE. It covers 53300 on one host, two-host DSNs that mix 53300, 28000 and a silent host in both orders, and sslmode=prefer.

Fixes #2278

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

Coverage Stats

10 of 10 new or added lines in 1 file covered. (100.0%)

6 existing lines in 3 files now uncovered.

30108 of 32939 relevant lines covered (91.41%)

1247.72 hits per line