-
Notifications
You must be signed in to change notification settings - Fork 30k
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
src: port coverage serialization to C++ #26874
Conversation
@joyeecheung fascinating, I'll provide a thorough code review sometime tomorrow (including checking out the branch and testing). Assuming that collecting coverage continues to work as expected for the Node.js codebase and user-land modules, this refactor seems like a no-brainer. Great work. |
7a88639
to
3768e82
Compare
I updated this patch to accommodate #26544 - we now must use cc @addaleax |
3768e82
to
14147bb
Compare
29b45ea
to
8a8d45a
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.
I'm running into some issues attempting to run with:
make coverage -j4
the testsequential/test-inspector-contexts
seems to hang indefinitely.
☝️ this seems to be happening to me on master too.
- running the
c8
reporting library seems to fail due to memory issues:
Security context: 0x298904f1a281 <JSObject>
0: builtin exit frame: parse(this=0x298904f0ab61 <Object map = 0x298980501449>,0x298977288139 <Very long string[204639]>,0x298904f0ab61 <Object map = 0x298980501449>)
1: /* anonymous */(aka /* anonymous */) [0x29893a080139] [/Users/bencoe/oss/node/node_modules/c8/lib/report.js:102] [bytecode=0x2989343531b9 offset=59](this=0x29896de004d1 <undefined>,0x29893a0e4ef1 <String[34]: coverage-8...
FATAL ERROR: Ineffective mark-compacts near heap limit Allocation failed - JavaScript heap out of memory
1: 0x100068dc2 node::Abort() [/Users/bencoe/oss/node/./node]
2: 0x10006943a node::OnFatalError(char const*, char const*) [/Users/bencoe/oss/node/./node]
are you bumping into this issue with make coverage
; I can try on my home computer as well (It's worth noting this also happened to me on master
).
std::unique_ptr<profiler::V8CoverageConnection> connection); | ||
profiler::V8CoverageConnection* coverage_connection(); | ||
|
||
inline void set_coverage_directory(const char* directory); |
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.
will the directory passed in to set_coverage_directory
have been resolved
? I think this could protect against some issues that crop up if the cwd
is changed by a subprocess.
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.
This is set as usual during pre-execution (but it's called from JS to persist states in C++ now), so yes it is resolved.
@bcoe The OOM is probably related to the heap V8 has allocated rather than physical memory size. Try a bigger heap e.g. |
I ran into this before, try cleaning up the |
This patch moves the serialization of coverage profiles into C++. With this we no longer need to patch `process.reallyExit` and hook into the exit events, but instead hook into relevant places in C++ which are safe from user manipulation. This also makes the code easier to reuse for other types of profiles.
8a8d45a
to
a63c7f2
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.
You were correct that the underlying issue was not cleaning up Release/.coverage
doh.
I've tested locally on yargs, and things seem to work better than ever 👍
Thanks for the hard work.
Landed in 864860e |
This patch moves the serialization of coverage profiles into C++. With this we no longer need to patch `process.reallyExit` and hook into the exit events, but instead hook into relevant places in C++ which are safe from user manipulation. This also makes the code easier to reuse for other types of profiles. PR-URL: #26874 Reviewed-By: Ben Coe <[email protected]>
// TODO(joyeecheung): white list the signals? | ||
|
||
#if HAVE_INSPECTOR | ||
profiler::EndStartedProfilers(env); |
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.
@joyeecheung Should this be guarded to only be run if sig > 0
and pid
is one of 0, -1, getpid(), -getpid()
or something similar?
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 think we should only write the profiles when we are sure the kill is initiated to terminate the process, are those conditions enough for us to be sure about that?
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.
@joyeecheung I mean, some signals that would usually terminate the process can be caught (like SIGINT
), so we cannot be sure of whether this is going to terminate the process or not no matter what…
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 we can only take care of signals that don't have a handler attached through Node.js's APIs, though even then it would get complicated because of workers. Or we could just not guard this at all, or let the user decide somehow, but then no one has reported any issues with this so I don't know what the real-world use cases would be.
This patch moves the serialization of coverage profiles into
C++. With this we no longer need to patch
process.reallyExit
and hook into the exit events, but instead hook into relevant
places in C++ which are safe from user manipulation. This also
makes the code easier to reuse for other types of profiles.
Checklist
make -j4 test
(UNIX), orvcbuild test
(Windows) passes