Stabilize GB300 nightly tests (#35044)

2026-08-17 03:30:16 -07:00
parent 744740dbea
commit 82995a001b
13 changed files with 515 additions and 351 deletions
@@ -0,0 +1,6 @@
+"""Shared configuration for tests on the dedicated GB300 runner."""
+
+# The runner's ephemeral client-port range is 10240-65535. Keep the TCPStore
+# rendezvous below it so a recently closed client connection cannot make the
+# subsequent TCPStore bind fail with EADDRINUSE.
+GB300_NCCL_PORT = "10000"