Details
- Reviewers
rmacklem - Group Reviewers
network - Commits
- rG033b4c1bff10: kgssapi: use netlink RPC client to talk to gssd(8)
Diff Detail
- Repository
- rG FreeBSD src repository
- Lint
Lint Not Applicable - Unit
Tests Not Applicable
Event Timeline
| sys/kgssapi/gss_impl.c | ||
|---|---|---|
| 92 | I'm unsure about the exact sequence of events expected with respect to starting this timeout and starting the daemon, but 1 minute isn't super long under emulation so I'm not sure if reducing from 5 to 1 minute is likely to be a problem. | |
| sys/kgssapi/gss_impl.c | ||
|---|---|---|
| 92 | I decided to change the timeout parameters together with the transport. With the unix/stream socket I really don't see much value in retries at all: if the daemon is absent, we would get error; if the daemon has hang, retries doesn't make sense - unix/stream is reliable, the data is waiting in the socket for the daemon. 5 minutes sounds like overkill to me. If you want I can get it back to 5 minutes, but I will do also increase retries then. So that client keeps sending the RPC call constantly. If the daemon is restarted, doesn't make sense to wait a minute. P.S. Thanks a lot for looking at the patches! | |
- Put back the original overall timeout of 5 minutes. Let's not mix this with change of transport.
- However, put much more retries into this timeout. The nature of the clnt_nl is different to clnt_reconnect + clnt_vc, and we should retry much more often.