Add an automated conhost benchmark tool#16453
Conversation
|
I'll leave this PR in draft until next year when we're under less time-to-ship-urgency. Since it's just a tool it's not important to review and/or merge it in anyways after all. |
|
I know it's not ready yet, but I wanna point out a weird condition I'm seeing when running two conhost binaries together. The second one to run always has a worse performance trace for some reason. traces
|
|
The latest commit should fix that issue. The old code did effectively The new code is a bit simpler (hopefully), but it also includes a better way to track time limits, etc. It should be a bit more robust now. It can still randomly hang though, until we merge #16457 (just press Ctrl+C if that happens and repeat it). There can also be quite some variation for the |
conhost has 2 bugs related to clipboard handling: * Missing retry on `OpenClipboard`: When copying to the clipboard explorer.exe is very eager to open the clipboard and peek into it. I'm not sure why it happens, but I can see `CFSDropTarget` in the call stack. It uses COM RPC and so this takes ~20ms every time. That breaks conhost's clipboard randomly during `ConsoleBench`. During non-benchmarks I expect this to break during RDP. * Missing null-terminator check during paste: `CF_UNICODETEXT` is documented to be a null-terminated string, which conhost v2 failed to handle as it relied entirely on `GlobalSize`. Additionally, this changeset simplifies the `HGLOBAL` code slightly by adding `_copyToClipboard` to abstract it away. ## Validation Steps Performed * `ConsoleBench` (#16453) doesn't fail randomly anymore ✅
conhost has 2 bugs related to clipboard handling: * Missing retry on `OpenClipboard`: When copying to the clipboard explorer.exe is very eager to open the clipboard and peek into it. I'm not sure why it happens, but I can see `CFSDropTarget` in the call stack. It uses COM RPC and so this takes ~20ms every time. That breaks conhost's clipboard randomly during `ConsoleBench`. During non-benchmarks I expect this to break during RDP. * Missing null-terminator check during paste: `CF_UNICODETEXT` is documented to be a null-terminated string, which conhost v2 failed to handle as it relied entirely on `GlobalSize`. Additionally, this changeset simplifies the `HGLOBAL` code slightly by adding `_copyToClipboard` to abstract it away. * `ConsoleBench` (#16453) doesn't fail randomly anymore ✅ (cherry picked from commit 86c30bd)
conhost has 2 bugs related to clipboard handling: * Missing retry on `OpenClipboard`: When copying to the clipboard explorer.exe is very eager to open the clipboard and peek into it. I'm not sure why it happens, but I can see `CFSDropTarget` in the call stack. It uses COM RPC and so this takes ~20ms every time. That breaks conhost's clipboard randomly during `ConsoleBench`. During non-benchmarks I expect this to break during RDP. * Missing null-terminator check during paste: `CF_UNICODETEXT` is documented to be a null-terminated string, which conhost v2 failed to handle as it relied entirely on `GlobalSize`. Additionally, this changeset simplifies the `HGLOBAL` code slightly by adding `_copyToClipboard` to abstract it away. * `ConsoleBench` (#16453) doesn't fail randomly anymore ✅ (cherry picked from commit 86c30bd)
OpenConsole.sln
Outdated
| {328729E9-6723-416E-9C98-951F1473BBE1}.Release|x64.ActiveCfg = Release|x64 | ||
| {328729E9-6723-416E-9C98-951F1473BBE1}.Release|x86.ActiveCfg = Release|Win32 | ||
| {BE92101C-04F8-48DA-99F0-E1F4F1D2DC48}.AuditMode|Any CPU.ActiveCfg = Debug|Win32 | ||
| {BE92101C-04F8-48DA-99F0-E1F4F1D2DC48}.AuditMode|ARM.ActiveCfg = Debug|Win32 |
There was a problem hiding this comment.
now you gotta drop all the ARM lines from this file!
I use vim:
:g/\<ARM\>/d
:wq
zadjii-msft
left a comment
zadjii-msft
left a comment
There was a problem hiding this comment.
oh man we so should run this and publish these as a CI artifact. that'd be neat
4 participants
ConsoleBenchis capable of launching conhost instances and measuringtheir performance characteristics. It writes these out as an HTML file
with violin graphs (using plotly.js) for easy comparison.
Currently, only a small number of tests have been added, but the code
is structured in such a way that more tests can be added easily.