[Request] Add option to trigger a heap dump

[lkonno]

Describe the feature

Add option to trigger the heap dump capture remotely and retrieve the dump file. Currently, it is necessary to execute the jmap command inside the pod.

It can be helpful when the image does not include the jmap/jcmd tools.

HotSpotDiagnosticMXBean

Anything other information?

[grzesuav]

Also interested in this feature, especially for container running with distroless images where there is no shell to even execute jmap/jcmd

[andrewazores]

We are targeting this for the next major release (4.0), however it's likely that it will require the installation of the Cryostat Agent into your target application(s) in order to transfer the heap dump out from the target. Otherwise, we'd be able to remotely trigger the heap dump, but the dump file would remain stuck in the container's local filesystem.

[andrewazores]

An interesting idea (needs proof of concept) would be to try creating a Unix domain socket:

https://inside.java/2021/02/03/jep380-unix-domain-sockets-channels/

https://github.com/kohlschutter/junixsocket

This way, /path/to/socket could possibly be given to the Diagnostic MXBean as the path to dump the hprof file. The Agent could be listening on this socket and piping data from the Unix socket to a network socket (HTTP connection) back to Cryostat. This way, we might be able to avoid writing to ephemeral container storage in tmpfs or similar, which could add some extra complexity (OOM or out-of-diskspace)