Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

fix(postgres): treat Supavisor's :einval socket drop as a transient retry (#126135)

Tom Owers720bd02aebfc
diff --git a/products/warehouse_sources/backend/temporal/data_imports/sources/postgres/postgres.py b/products/warehouse_sources/backend/temporal/data_imports/sources/postgres/postgres.py
index 852c6b7af60e..802263936b0c 100644
--- a/products/warehouse_sources/backend/temporal/data_imports/sources/postgres/postgres.py
+++ b/products/warehouse_sources/backend/temporal/data_imports/sources/postgres/postgres.py
@@ -437,6 +437,11 @@ _CONNECTION_DROPPED_ERROR_SUBSTRINGS = (
# "(authenticated)" wrapper would be too broad: a non-:closed "Internal error (authenticated): ..."
# could be a permanent pooler/protocol failure that should surface immediately, not be retried.
#
+# ":einval" is the same shape with a different Erlang reason: gen_tcp returns it (POSIX EINVAL,
+# "invalid argument") when a socket operation is attempted on a connection the port has already
+# torn down — the same dead-backend-socket condition as ":closed", just raised on whichever side of
+# the race loses after the peer closes. A fresh reconnect recovers the same way.
+#
# Neon's own compute-side WAL relay ("walsender") surfaces the same generic XX000 InternalError_
# when it loses connectivity to a safekeeper — the storage-tier peer that actually holds the WAL,
# since a Neon compute doesn't keep it locally: "[walsender] Failed to read WAL (...): failed to
@@ -449,6 +454,7 @@ _POOLER_CONNECTION_DROPPED_ERROR_SUBSTRINGS = (
"echeckoutretries",
"echeckouttimeout",
"internal error (authenticated): :closed",
+ "internal error (authenticated): :einval",
"failed to connect to safekeeper",
)
diff --git a/products/warehouse_sources/backend/temporal/data_imports/sources/postgres/test_postgres.py b/products/warehouse_sources/backend/temporal/data_imports/sources/postgres/test_postgres.py
index 6fb37afbb5a6..f867ce66fdfd 100644
--- a/products/warehouse_sources/backend/temporal/data_imports/sources/postgres/test_postgres.py
+++ b/products/warehouse_sources/backend/temporal/data_imports/sources/postgres/test_postgres.py
@@ -2733,6 +2733,10 @@ class TestIsConnectionDroppedError:
# error code, so it's matched on the full phrase including the ":closed" reason; same
# transient class as the pooler drops above and recovers on reconnect.
psycopg.errors.InternalError_("Internal error (authenticated): :closed"),
+ # Same Supavisor pooler-drop shape, but raised with Erlang's ":einval" (POSIX EINVAL)
+ # reason instead of ":closed" — gen_tcp returns it when an operation hits a backend
+ # socket the pooler has already torn down. Same transient class; recovers on reconnect.
+ psycopg.errors.InternalError_("Internal error (authenticated): :einval"),
# Supavisor reports a transient timeout reaching the upstream backend as a
# ConnectionFailure (08006, an OperationalError) carrying the Erlang-tuple reason
# "{:error, :etimedout}" — a transient drop the in-process recovery must catch.