Use TCP keepalive to detect broken sockets #681
Conversation
vsomeip does not enforce a single socket between endpoints; it does not actively close an existing socket after a new one is established. It is possible that: - socket is established from nodeA to nodeB - ethernet gets unplugged - nodeA tries to transmit, times out, closes the socket - ethernet is reconnected - nodeA establishes a new socket to node B nodeB never received an RST while the cable was unplugged, still thinks the old socket is established, and effectively "leaks" until TCP keepalive detects the break. At default values this takes 2 hr 10 min. This can be mitigated by reducing the keepalive settings. The implementation of TCP keepalive is different between OS: - Linux can configure all 3 settings on a per-socket basis - QNX can configure 1 setting per-socket and 2 settings globally QNX would need to set the global values using sysctl, perhaps at startup.
|
@joeyoravec is this ready to review or are you still working on this? |
I've left this in draft so it's clear that this is not ready for merge. Yes this "works" and is worth reviewing. |
| if (rc != 0) { | ||
| VSOMEIP_WARNING << "tcp_server_endpoint::connect: couldn't enable keep_alive(10)" ; | ||
| } | ||
| #ifndef __QNX__ |
There was a problem hiding this comment.
Hey @joeyoravec
Thanks for this Pull Request.
One suggestion to maybe avoid "#ifdef platform" and the hardcoded values, the tcp keep alive options could be added as new configurations in the vsomeip json files.
If no configurations would be present, the default from the platform would be used (this would be preferable for linux based systems, where sysctl is used)
If configurations for the keep alive would be present, those would be used instead, for all platforms.
fixes #674
vsomeip does not enforce a single socket between endpoints; it does not actively close an existing socket after a new one is established.
It is possible that:
nodeB never received an RST while the cable was unplugged, still thinks the old socket is established, and effectively "leaks" until TCP keepalive detects the break. At default values this takes 2 hr 10 min. This can be mitigated by reducing the keepalive settings.
The implementation of TCP keepalive is different between OS:
QNX would need to set the global values using sysctl, perhaps at startup.