-
Notifications
You must be signed in to change notification settings - Fork 29.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
inspector: added NodeRuntime domain #27600
Conversation
74a3c97
to
07f50b5
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for working on this!
src/inspector_agent.cc
Outdated
@@ -779,7 +797,8 @@ void Agent::WaitForDisconnect() { | |||
} | |||
// TODO(addaleax): Maybe this should use an at-exit hook for the Environment | |||
// or something similar? | |||
client_->contextDestroyed(parent_env_->context()); | |||
if (!client_->reportWaitingForDebuggerToDisconnect()) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe put this to NodeInspectorClient::contextDestroyed?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'd like to have it here, otherwise in future someone might use contextDestroyed in other places and get misleading notification.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Feel free to rename "contextDestroyed" to something else - like "mainContextDone"?
2e3c872
to
051a90c
Compare
src/inspector/node_protocol.pdl
Outdated
experimental domain NodeRuntime | ||
# When domain is enabled `NodeRuntime.waitingForDisconnect` will be issued at the end of node process. | ||
# Please see `NodeRuntime.waitingForDisconnect` for more details. | ||
command enable |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
"enable" seems to generic a verb for this very specific task. Maybe "retainMainContextForDebugger"?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Renamed to notifyWhenWaitingForDisconnect.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Did you forget to push the change?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
it looks like something went wrong yesterday, I just pushed updated version.
src/inspector_agent.cc
Outdated
@@ -779,7 +797,8 @@ void Agent::WaitForDisconnect() { | |||
} | |||
// TODO(addaleax): Maybe this should use an at-exit hook for the Environment | |||
// or something similar? | |||
client_->contextDestroyed(parent_env_->context()); | |||
if (!client_->reportWaitingForDebuggerToDisconnect()) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Feel free to rename "contextDestroyed" to something else - like "mainContextDone"?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for working on this!
051a90c
to
020da49
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@eugeneo please take another look.
020da49
to
1a3f130
Compare
# Event is issued when Node process is finished and waiting for debugger to disconnect. | ||
# When enabled client should listen for this event instead of `Runtime.executionContextDestroyed` | ||
# to check wether node process is finished. | ||
event waitingForDisconnect |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I believe it needs to be communecated clearly on the purpose of this notification vs the Runtime.executionContextDestroyed
- specifically, that the main context is retained and can still be used.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I updated comment.
@@ -610,6 +620,14 @@ class NodeInspectorClient : public V8InspectorClient { | |||
return false; | |||
} | |||
|
|||
bool notifyWaitingForDisconnect() { | |||
for (const auto& id_channel : channels_) { | |||
if (id_channel.second->notifyWaitingForDisconnect()) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It looks like this notification will only get sent to the sessions that requested it. Others will get nothing, not even an Runtime.executionContextDestroyed
.
I suggest sending other sessions executionContextDestroyed
either as al else statement for this if or at a later point, once all sessions that requested a disconnect notification disconnect.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It is not possible to send executionContextDestroyed
without dirty hacks. We need to intercept every Runtime.enable
and Runtime.disable
calls to maintain runtime enabled state on node side. At the same time we can not just calls executionContextDestroyed on inspector since then inspector will mark context as destroyed and any later calls that uses Runtime will return an error.
Based on how much hacks we need to add, how many clients there are in a wild right now, what is probability of using new client and old client at the same time I believe that we can just send nothing here. Protocol definition stated exactly this.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I tried to adopt solution that you proposed and it seems like it works!
bbdcbc8
to
0c5cb42
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks!
Looks like this PR needs to be rebased by hand. |
If you are having CI difficulties, the actual problem is almost certainly the merge-commit. Our Jenkins setup often (always?) does not handle merge-commits well. Squash the merge-commit into the other commit or force-push the merge-commit out, and then CI should be fine. |
c7c45b4
to
0c5cb42
Compare
0c5cb42
to
6ecbb2c
Compare
Historically Node process sends Runtime.executionContextDestroyed with main context as argument when it is finished. This approach has some disadvantages. V8 prevents running some protocol command on destroyed contexts, e.g. Runtime.evaluate will return an error or Debugger.enable won't return a list of scripts. Both command might be useful for different tools, e.g. tool runs Profiler.startPreciseCoverage and at the end of node process it would like to get list of all scripts to match data to source code. Or some tooling frontend would like to provide capabilities to run commands in console when node process is finished to allow user to inspect state of the program at exit. This PR adds new domain: NodeRuntime. After NodeRuntime.notifyWhenWaitingForDisconnect is enabled by at least one client, node will send NodeRuntime.waitingForDebuggerToDisconnect event instead of Runtime.executionContextDestroyed. Based on this signal any protocol client can capture all required information and then disconnect its session.
6ecbb2c
to
d53eb62
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Test LGTM. Rest of it is RSLGTM/defer-to-@eugeneo-and-others.
Landed in 10d7e01
|
Historically Node process sends Runtime.executionContextDestroyed with main context as argument when it is finished. This approach has some disadvantages. V8 prevents running some protocol command on destroyed contexts, e.g. Runtime.evaluate will return an error or Debugger.enable won't return a list of scripts. Both command might be useful for different tools, e.g. tool runs Profiler.startPreciseCoverage and at the end of node process it would like to get list of all scripts to match data to source code. Or some tooling frontend would like to provide capabilities to run commands in console when node process is finished to allow user to inspect state of the program at exit. This PR adds new domain: NodeRuntime. After NodeRuntime.notifyWhenWaitingForDisconnect is enabled by at least one client, node will send NodeRuntime.waitingForDebuggerToDisconnect event instead of Runtime.executionContextDestroyed. Based on this signal any protocol client can capture all required information and then disconnect its session. PR-URL: #27600 Reviewed-By: Eugene Ostroukhov <eostroukhov@google.com> Reviewed-By: Rich Trott <rtrott@gmail.com>
Historically Node process sends Runtime.executionContextDestroyed with main context as argument when it is finished. This approach has some disadvantages. V8 prevents running some protocol command on destroyed contexts, e.g. Runtime.evaluate will return an error or Debugger.enable won't return a list of scripts. Both command might be useful for different tools, e.g. tool runs Profiler.startPreciseCoverage and at the end of node process it would like to get list of all scripts to match data to source code. Or some tooling frontend would like to provide capabilities to run commands in console when node process is finished to allow user to inspect state of the program at exit. This PR adds new domain: NodeRuntime. After NodeRuntime.notifyWhenWaitingForDisconnect is enabled by at least one client, node will send NodeRuntime.waitingForDebuggerToDisconnect event instead of Runtime.executionContextDestroyed. Based on this signal any protocol client can capture all required information and then disconnect its session. PR-URL: #27600 Reviewed-By: Eugene Ostroukhov <eostroukhov@google.com> Reviewed-By: Rich Trott <rtrott@gmail.com>
Historically Node process sends Runtime.executionContextDestroyed
with main context as argument when it is finished.
This approach has some disadvantages. V8 prevents running some
protocol command on destroyed contexts, e.g. Runtime.evaluate
will return an error or Debugger.enable won't return a list of
scripts.
Both command might be useful for different tools, e.g. tool runs
Profiler.startPreciseCoverage and at the end of node process it
would like to get list of all scripts to match data to source code.
Or some tooling frontend would like to provide capabilities to run
commands in console when node process is finished to allow user to
inspect state of the program at exit.
This PR adds new domain: NodeRuntime. This domain sends
NodeRuntime.waitingForDisconnect
when this event is enabled.Checklist
make -j4 test
(UNIX), orvcbuild test
(Windows) passes