Debugger IPC creation failure should not abort coreclr startup #90161
Conversation
|
Tagging subscribers to this area: @tommcdon Issue DetailsOn Linux, the debugger creates 2 single-directional pipes in We found that there are two places where the runtime creates pipes/domain sockets for diagnostics:
This PR changes Debugger Port creation failure from fatal to non-fatal.
|
| @@ -1939,7 +1939,8 @@ HRESULT Debugger::Startup(void) | |||
| if (FAILED(hr)) | |||
| { | |||
| ShutdownTransport(); | |||
| ThrowHR(hr); | |||
| STRESS_LOG0(LF_CORDB, LL_ERROR, "D::S: The debugger pipe failed to initialize in /tmp or $TMPDIR.\n"); | |||
| return S_OK; // we do not want debugger IPC to block runtime initialization | |||
There was a problem hiding this comment.
Don't know if we should return s_ok in those cases. Given other paths do it, maybe ok for .net 8. We should have other values to tear down the RC thread.
There was a problem hiding this comment.
We need to track properly cleaning up (returning S_FALSE here and if the env var is set so the caller does the cleanup) for .NET 9.
There was a problem hiding this comment.
I'll add returning S_FALSE. As Mike points out this will ensure proper cleanup:
https://github.com/tommcdon/runtime/blob/c9b4d5e7ad45de9ef4181cac45439cdc8d60033c/src/coreclr/vm/ceemain.cpp#L1833-L1843
There was a problem hiding this comment.
I reverted the change and will add the S_FALSE implementation in .NET 9
tommcdon
force-pushed
the
dev/tommcdon/graceDebuggerInitFail
branch
from
79cfeb5 to
c9b4d5e
Compare
ghost
locked as resolved and limited conversation to collaborators
On Linux, the debugger creates 2 single-directional pipes in
/tmpor$TMPDIR. Today if we fail to create these pipes, coreclr init will fail withFailed to create CoreCLR, HRESULT: 0x8007000Eas described in #86259.We found that there are two places where the runtime creates pipes/domain sockets for diagnostics:
$TMPDIRif the env variable is set. It provides a transport for collecting memory dumps, starting an EventPipe tracing session, requesting a dump, ... It is considered non-fatal if it doesn't start.$TMPDIRif the variable is set in Debugger::Startup. This is considered fatal if we can't create the port. This is not only related to read only file systems or non-Linux file system mounts. For example, if$TMPDIRis set to a Windows mount in WSL2, the pipe will also fail to be created.This PR changes Debugger Port creation failure from fatal to non-fatal.