Project

General

Profile

Actions

Bug #39249

open

Recover from false registration failures caused by TCP reset at high concurrency

Added by Pablo Méndez Hernández 3 months ago. Updated 3 months ago.

Status:
Ready For Testing
Priority:
Normal
Assignee:
-
Category:
-
Target version:
-
Difficulty:
Triaged:
No
Fixed in Releases:
Found in Releases:

Description

At 320+ concurrent host registrations, the server sometimes creates the Candlepin consumer record successfully but the TCP connection is reset before the response reaches subscription-manager. This causes subscription-manager to report failure even though the registration succeeded server-side.

Adds a recovery check after subscription-manager register fails: runs subscription-manager identity to detect whether the host was actually registered, and if so treats the registration as successful rather than triggering a retry loop.

Actions

Also available in: Atom PDF