| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
| |
We don't need to lookup the addresses every
time we get btime. Do it once during hbeat_init
and reuse the socket in hbeat. This cleans up
the qarsh strace so the hbeat is only a send and recv.
|
|
|
|
| |
I don't see any way to coexist with the old "protocol"
|
| |
|
|
|
|
|
| |
A quarter of a second may be too short for some connections.
Increasing it to half a second should make it more reliable.
|
| |
|
|
|
|
|
|
| |
of recvfroms and sleeps in btime(). This makes it easier
to tune how long to wait for a heartbeat. Now I can get
heartbeats over my vpn connection.
|
| |
|
|
|
|
|
|
|
| |
waiting around long enough for the heartbeat response to come back.
Since we only heartbeat when there isn't any output (in qarsh),
taking up to a second won't hurt. After doing some tests I'm changing
the recvfrom to recvfrom sleep to 1000 usec and retrying up to 40 times.
|
| |
|
|
|
|
|
|
|
|
|
| |
Two factors forced this change.
1. ports were being re-used so quickly that we could get lingering
packets from past hosts when getting the btime for a new host.
2. We can't control which IP a btime response is sent from if a system
has more than one interface. Adding a cookie allows us to know that
a response came from whom we sent it.
|
| |
|
|
|