Add ssh keepalives to ssh communicator #19339
Conversation
There was a problem hiding this comment.
Hi @spion06,
Thanks for the contribution!
While adding keepalive messages might be useful in some circumstances, I'm wondering if it's a knob we need to expose to users at all. The provisioner are not long-lived connections, so the tuning of the frequency isn't really of much use. I think we might just want a sane default (less than 30s, since that's a common network timeout setting elsewhere) and always have it enabled.
Hold off on rewriting the PR though, until we have a plan for implementation.
| // Start a goroutine that executes every KeepAliveInterval to keep the session alive | ||
| // will return when the session no longer responds to requests | ||
| go func() { | ||
| if c.config.KeepAliveInterval < 0 { |
There was a problem hiding this comment.
This will panic below with a timeout of 0, and should probably be limited to a reasonable minimum like 1 second, so we don't flood the connection with 1ns keepalives.
| @@ -51,6 +54,9 @@ type connectionInfo struct { | |||
| ScriptPath string `mapstructure:"script_path"` | |||
| TimeoutVal time.Duration `mapstructure:"-"` | |||
|
|
|||
| KeepAliveInterval string `mapstructure:"keep_alive_interval"` | |||
There was a problem hiding this comment.
This isn't added to config validation, so it can't actually be used from a configuration.
|
Closing in favor of #20437, which removes the config option and only starts one keepalive routine per connection. |
|
I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues. If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further. |
ghost
locked and limited conversation to collaborators
This helps resolve an issue in stigged environments where the server-side ClientAliveInterval is set to a lower value (such as 300) and ClientAliveCount is set to 0. During the ssh provisioner if there is a long pause without data the server-side will disconnect causing terraform to fail. This would resolve that issue by sending session-level keepalive requests. The overhead of the ssh keepalive is minimal and shouldn't impact performance of anything.
This should resolve #18517 based off a cursory search of open issues.