Bug 180477 – We should be able to recover after a Service Worker process crash

• Component: WebKit2
• Status: RESOLVED FIXED

[Chris Dumez]

We should be able to recover after a Service Worker process crash.

[Chris Dumez]

Created attachment 328654 [details]
WIP Patch

[Chris Dumez]

Created attachment 328656 [details]
WIP Patch

Will clean up tonight.

[Chris Dumez]

*** Bug 180496 has been marked as a duplicate of this bug. ***

[Chris Dumez]

Created attachment 328675 [details]
Patch

[Brady Eidson]

Comment on attachment 328675 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328675&action=review

> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.h:98
> +    template<typename U> static void sendToContextProcess(U&& message);

Why static?

In a future world of multiple context processes, how can this possibly be static?

If it's a convenience because of the lambda captures... can we capture the connection, or a way of looking the connection up?

[Chris Dumez]

Comment on attachment 328675 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328675&action=review

>> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.h:98
>> +    template<typename U> static void sendToContextProcess(U&& message);
> 
> Why static?
> 
> In a future world of multiple context processes, how can this possibly be static?
> 
> If it's a convenience because of the lambda captures... can we capture the connection, or a way of looking the connection up?

Yes, it is because of the lambda captures. It would be unsafe to capture |this| was I have no way to make sure the connection is still alive when the lambda is called. I'll try and find another way without making this static.

[Chris Dumez]

Comment on attachment 328675 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328675&action=review

>>> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.h:98
>>> +    template<typename U> static void sendToContextProcess(U&& message);
>> 
>> Why static?
>> 
>> In a future world of multiple context processes, how can this possibly be static?
>> 
>> If it's a convenience because of the lambda captures... can we capture the connection, or a way of looking the connection up?
> 
> Yes, it is because of the lambda captures. It would be unsafe to capture |this| was I have no way to make sure the connection is still alive when the lambda is called. I'll try and find another way without making this static.

WebSWServerConnection is not refcounted so I cannot protect it when I capture.

[youenn fablet]

Comment on attachment 328675 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328675&action=review

> Source/WebCore/ChangeLog:12
> +        Once the connection is the context is established, process "run service worker"

s/is/in

> Source/WebCore/ChangeLog:17
> +        if processed. We used to assert that we had a connection to the context process.

s/if/is

> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.cpp:123
> +        server().runServiceWorkerIfNecessary(*serviceWorkerIdentifier, [connectionIdentifier = identifier(), fetchIdentifier, serviceWorkerIdentifier, request, options, formData](bool success) {

We should probably have WebSWServerConnection::startFetch take rvalue references for request, options and formData to take benefit of these here.
Space between ] ( although I am not a fan either.

Ditto in below WebSWServerConnection::XX methods.

> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.cpp:125
> +                sendToContextProcess(Messages::WebSWContextManagerConnection::StartFetch { connectionIdentifier, fetchIdentifier, serviceWorkerIdentifier, request, options, formData });

In case success is false, we need to return a DidNotHandle message on the web content process.

> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.h:55
> +    IPC::Connection& ipcConnection() const { return m_contentConnection.get(); }

Not sure ipcConnection is a great name since there is also a service worker process connection.
Should it be something like contentConnection() or webContentConnection() instead?

> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.h:101
>      Ref<IPC::Connection> m_contentConnection;

Hum, m_contextConnection is never used so we could remove it, or start using it.
Probably better to just remove it for now. We are probably not ready to split service workers into separate processes based on origins.

> Source/WebKit/StorageProcess/StorageProcess.cpp:112
> +    // Relaunch the context process if necessary.

Is this comment needed?

> Source/WebKit/StorageProcess/StorageProcess.cpp:124
> +    if (m_swServerConnections.size() == 1 && m_serverToContextConnection && &m_swServerConnections.begin()->value->ipcConnection() == m_serverToContextConnection->ipcConnection())

Instead, should we have WebSWServerConnection has a bool isServiceWorkerProcessConnection()?

> LayoutTests/http/tests/workers/service/resources/postmessage-after-sw-process-crash.js:19
> +        }, 0);

Are we sure 0 is sufficient to ensure it will crash before the messages?
Should we not try to make the test a little bit more robust with something like:
- a first message to the SW to set a value inside the worker, the SW having a default value.
- a second message to get from the SW the stored value, which should be the default value if the worker process was crashed and the service worker rerun.

[Chris Dumez]

Comment on attachment 328675 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328675&action=review

>> LayoutTests/http/tests/workers/service/resources/postmessage-after-sw-process-crash.js:19
>> +        }, 0);
> 
> Are we sure 0 is sufficient to ensure it will crash before the messages?
> Should we not try to make the test a little bit more robust with something like:
> - a first message to the SW to set a value inside the worker, the SW having a default value.
> - a second message to get from the SW the stored value, which should be the default value if the worker process was crashed and the service worker rerun.

I am sure. See https://bugs.webkit.org/show_bug.cgi?id=180496 where the test was failing before I implemented recovery.

[Chris Dumez]

Comment on attachment 328675 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=328675&action=review

>> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.h:55
>> +    IPC::Connection& ipcConnection() const { return m_contentConnection.get(); }
> 
> Not sure ipcConnection is a great name since there is also a service worker process connection.
> Should it be something like contentConnection() or webContentConnection() instead?

Just trying to be consistent with Brady here who used WebSWServerToContextConnection::ipcConnection().

[Brady Eidson]

(In reply to Chris Dumez from comment #10)
> Comment on attachment 328675 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=328675&action=review
> 
> >> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.h:55
> >> +    IPC::Connection& ipcConnection() const { return m_contentConnection.get(); }
> > 
> > Not sure ipcConnection is a great name since there is also a service worker process connection.
> > Should it be something like contentConnection() or webContentConnection() instead?
> 
> Just trying to be consistent with Brady here who used
> WebSWServerToContextConnection::ipcConnection().

Meh - Use something more sensical if such a thing exists.

[Chris Dumez]

(In reply to Brady Eidson from comment #11)
> (In reply to Chris Dumez from comment #10)
> > Comment on attachment 328675 [details]
> > Patch
> > 
> > View in context:
> > https://bugs.webkit.org/attachment.cgi?id=328675&action=review
> > 
> > >> Source/WebKit/StorageProcess/ServiceWorker/WebSWServerConnection.h:55
> > >> +    IPC::Connection& ipcConnection() const { return m_contentConnection.get(); }
> > > 
> > > Not sure ipcConnection is a great name since there is also a service worker process connection.
> > > Should it be something like contentConnection() or webContentConnection() instead?
> > 
> > Just trying to be consistent with Brady here who used
> > WebSWServerToContextConnection::ipcConnection().
> 
> Meh - Use something more sensical if such a thing exists.

We have so many "connections", I am personally fine with ipcConnection() as it makes it clear it is the underlying IPC::Connection. contentConnection() or webContentConnection() are super vague and I would have no idea what they return from their name.

[Chris Dumez]

Created attachment 328679 [details]
Patch

[WebKit Commit Bot]

Comment on attachment 328679 [details]
Patch

Rejecting attachment 328679 [details] from commit-queue.

Failed to run "['/Volumes/Data/EWS/WebKit/Tools/Scripts/webkit-patch', '--status-host=webkit-queues.webkit.org', '--bot-id=webkit-cq-02', 'validate-changelog', '--check-oops', '--non-interactive', 328679, '--port=mac']" exit_code: 1 cwd: /Volumes/Data/EWS/WebKit

ChangeLog entry in LayoutTests/ChangeLog contains OOPS!.

Full output: http://webkit-queues.webkit.org/results/5525882

[Chris Dumez]

Created attachment 328680 [details]
Patch

[WebKit Commit Bot]

Comment on attachment 328680 [details]
Patch

Rejecting attachment 328680 [details] from commit-queue.

Failed to run "['/Volumes/Data/EWS/WebKit/Tools/Scripts/webkit-patch', '--status-host=webkit-queues.webkit.org', '--bot-id=webkit-cq-02', 'build', '--no-clean', '--no-update', '--build-style=release', '--port=mac']" exit_code: 2 cwd: /Volumes/Data/EWS/WebKit

Last 500 characters of output:
unsigned long long, std::optional<WTF::ObjectIdentifier<WebCore::ServiceWorkerIdentifierType> >, WebCore::ResourceRequest&&, WebCore::FetchOptions&&, IPC::FormDataReference&&) in WebSWServerConnection.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

** BUILD FAILED **


The following build commands failed:
	Ld /Volumes/Data/EWS/WebKit/WebKitBuild/Release/WebKit.framework/Versions/A/WebKit normal x86_64
(1 failure)

Full output: http://webkit-queues.webkit.org/results/5526070

[Chris Dumez]

Created attachment 328681 [details]
Patch

[youenn fablet]

But do we really need this ipcConnection() if it is only used to know which server connection we are talking to, in which case a bool getter would be better?
Also ipcConnection may return a const ref if possible.

[Chris Dumez]

(In reply to youenn fablet from comment #18)
> But do we really need this ipcConnection() if it is only used to know which
> server connection we are talking to, in which case a bool getter would be
> better?
> Also ipcConnection may return a const ref if possible.

We do not have this information at the moment so adding the bool getter was not trivial. I can look into this in a follow up.

[WebKit Commit Bot]

Comment on attachment 328681 [details]
Patch

Clearing flags on attachment: 328681

Committed r225622: <https://trac.webkit.org/changeset/225622>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/35902914>