[LU-19517] lctl and lnet ping both use a default timeout that is too small and fixed

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Affects Version/s: Lustre 2.16.0
• Labels: llnl llnlfixready
• Severity: 3
• Assignee: Olaf Faaland
• Reporter: Olaf Faaland

Description

Both "lctl ping" and "lnetctl ping" always send a positive timeout value to the kernel when issuing a ping, and the default value is 1 sec.  This has a few issues:

• On our large fabrics (e.g. el capitan) pings time out long before the fabric itself has given up transmitting the ping message, so the user gets a "ping failed" when in fact it's still in process, unless they specify the timeout option.
• This prevents the kernel from using the default timeout based on DEFAULT_PEER_TIMEOUT
• It's different than what lnetctl(8) says the default is