[squeak-dev] The Inbox: WebClient-Core-cbc.118.mcz
Levente Uzonyi
leves at caesar.elte.hu
Mon Oct 29 20:54:52 UTC 2018
On Sun, 28 Oct 2018, Chris Cunningham wrote:
>
>
> On Sun, Oct 28, 2018, 17:55 Levente Uzonyi <leves at caesar.elte.hu> wrote:
> Hi Chris,
>
> On Sun, 28 Oct 2018, Chris Cunningham wrote:
>
> > Hi.
> > I was loading the VMMaker package(s), and after manually opening the debugger and restarting at #httpGet:do: about 10 times, I implemented this hack so that I didn't have to do that
> anymore.
> >
> > I *think* this is fixing the issue - haven't had it raise errors while 'timing out' on loading packages since this (the timeout were sub-second - the connection hadn't gone through
> yet). Still, it
> > might just be timing - this isn't really a repeatable bug.
>
> I'm sure this change helps with that issue, but it has unwelcome side
> effects to WebClient's other users.
>
> This this definitely will not be going to trunk.
>
> The real solution would be to fix the server.
>
> Where did you see sub-second timeouts? The default timeout should be 45
> seconds.
>
> It didn't wait 45 seconds - it is almost instantaneous for me. The error received back (from Socket>>sendSomeData:startIndex:count:for: ) is "ConnectionTimedOut: send data timeout; data not sent",
> but I'm pretty darn certain it is that the socket isn't yet connected (trace put into Socket>>waitForSendDoneFor: confirms this). Looking at #waitForSendDoneFor: shows that before any wait, it checks
> if the socket is connected - if not, it immediately exits with false, which triggers the time out error in the caller.
>
> If I trap it and immediately resend, then it works. Weird.
I've never seen that happening. Do you have a stack trace of the error?
Levente
>
> -cbc
>
> >
> > Not in Trunk because it is definitely a hack - but it makes things work nicer.
> >
> > Also, committing packages to the inbox with this loaded doesn't result in walkbacks (from timeouts and whatnot) for me. Although it does take a long time to finish.
>
> Uploads use PUT requests, so expect to still see walkbacks there.
>
> Levente
>
> >
> > -cbc
> >
> > On Sun, Oct 28, 2018 at 5:08 PM <commits at source.squeak.org> wrote:
> > A new version of WebClient-Core was added to project The Inbox:
> > http://source.squeak.org/inbox/WebClient-Core-cbc.118.mcz
> >
> > ==================== Summary ====================
> >
> > Name: WebClient-Core-cbc.118
> > Author: cbc
> > Time: 28 October 2018, 5:08:23.571079 pm
> > UUID: 683fbe3b-418f-a443-9a20-3f2a7af4b7e1
> > Ancestors: WebClient-Core-pre.117
> >
> > A hack to work around connectionTimedOut annoyances when opening packages from Trunk (sometimes).
> >
> > =============== Diff against WebClient-Core-pre.117 ===============
> >
> > Item was changed:
> > ----- Method: WebClient>>httpGet:do: (in category 'methods') -----
> > httpGet: urlString do: aBlock
> > "GET the response from the given url"
> > "(WebClient httpGet: 'http://www.squeak.org') content"
> >
> > + | request errCount |
> > - | request |
> > self initializeFromUrl: urlString.
> > request := self requestWithUrl: urlString.
> > request method: 'GET'.
> > userAgent ifNotNil:[:ua | request headerAt: 'User-Agent' put: ua].
> > self contentDecoders ifNotNil: [:decoders | request headerAt: 'Accept-Encoding' put: decoders].
> > +
> > + errCount := 0. "Let's try resending to get around 'connection issues' trunk connections"
> > + [
> > + aBlock value: request.
> > + ^self sendRequest: request
> > + ] on: Error, NetworkError do: [:e| debugLog ifNotNil: [debugLog cr; nextPutAll: 'httpGet error: ', e; flush]. (errCount := errCount + 1) > 3 ifTrue: [e outer]. e retry].!
> > - aBlock value: request.
> > - ^self sendRequest: request
> > - !
> >
> >
> >
> >
>
>
>
More information about the Squeak-dev
mailing list
|