717759 - negative pipelining feedback on failed image parses

Status: RESOLVED FIXED

(Core :: Networking: HTTP, defect)

Description

[Patrick McManus [:mcmanus]]

this is a long planned item for pipelining "negative feedback". Correlated with the old pipelining failures were pages with a lot of broken images (occasionally js or css was getting fed into images) - for various reasons most of that has disappeared from the net, but its still a reasonable metric for whether pipelining is running ok - at least on the negative side.

Even if there are legit errors that means cancellations are going to happen, connections are going to close, and that's bad for pipelining too. Basically, pipelines are speculating on a future clean environment.

Consumer code that discovers a problem with data from necko can report its nsiuri through the new net:failed-to-process-uri observer event. Right now the only thing that makes that kind of report is imagelib - it seemed to me that parse errors are just too common in other kinds of resources. Right now necko just uses this as pipeline feedback, but you can imagine maybe invalidating caches feeding this information into a database as potential future uses of it.

there's even a test.

Comment 1

[Patrick McManus [:mcmanus]]

Attachment: patch 0

Comment 2

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 588186 [details] [diff] [review]
patch 0

Looks fine to me, but could use an actual imagelib peer too.

Comment 3

[Patrick McManus [:mcmanus]]

Comment on attachment 588186 [details] [diff] [review]
patch 0

joe, can you glance at the imagelib bit of this?

Comment 4

[Joe Drew (not getting mail)]

Comment on attachment 588186 [details] [diff] [review]
patch 0

Review of attachment 588186 [details] [diff] [review]:
-----------------------------------------------------------------

One thing to note is that we're only guaranteed to look at the first N bytes of an image (N changes as of bug 715308), so you cannot rely on this to guarantee that images decode correctly. That is, onerror only fires if the file is absolutely not an image, not if it is subtly corrupt.

::: image/src/imgRequest.cpp
@@ +704,5 @@
> +
> +    nsCOMPtr<nsIObserverService> os = mozilla::services::GetObserverService();
> +    if (os)
> +      os->NotifyObservers(mURI, "net:failed-to-process-uri", nsnull);
> +  }

Please put this block above the comment block.

::: image/test/mochitest/test_net_failedtoprocess.html
@@ +30,5 @@
> +
> +  observe: function(subject, topic, data) {
> +    ok(topic == "net:failed-to-process-uri", "wrong topic");
> +    subject = subject.QueryInterface(Ci.nsIURI);
> +    ok(subject.asciiSpec == "chrome://mochitests/content/chrome/image/test/mochitest/invalid.jpg", "wong subject");

wrong, not wong

@@ +37,5 @@
> +};
> +
> +var obs = Cc["@mozilla.org/observer-service;1"].getService();
> +obs = obs.QueryInterface(Ci.nsIObserverService);
> +obs.addObserver(observer, "net:failed-to-process-uri", false);

I don't *think* this will work in a world where we need SpecialPowers to use chrome things. I'd prefer this test to be rewritten using SpecialPowers. (If this, in fact, is not just a chrome test; it's not obvious to me.)

Comment 5

[Honza Bambas (:mayhemer)]

Comment on attachment 588186 [details] [diff] [review]
patch 0

Review of attachment 588186 [details] [diff] [review]:
-----------------------------------------------------------------

::: image/src/imgRequest.cpp
@@ +703,5 @@
> +    // Report the URI to net:failed-to-decode-uri observers.
> +
> +    nsCOMPtr<nsIObserverService> os = mozilla::services::GetObserverService();
> +    if (os)
> +      os->NotifyObservers(mURI, "net:failed-to-process-uri", nsnull);

Hmmm... shouldn't this topic be more specific like:
- "net:failed-image-content-processing-on-uri"
- "net:image-content-corrupted", or at least
- "net:failed-to-process-uri-content"

This seems to me too general and could potentially collide, but up to you.

::: netwerk/protocol/http/nsHttpConnectionMgr.cpp
@@ +987,5 @@
> +    nsCAutoString host;
> +    PRInt32 port = -1;
> +    bool usingSSL = false;
> +
> +    nsresult rv = uri->SchemeIs("https", &usingSSL);

What if this is not http at all?

@@ +998,5 @@
> +
> +    nsRefPtr<nsHttpConnectionInfo> ci =
> +        new nsHttpConnectionInfo(host, port, nsnull, usingSSL);
> +    
> +    PipelineFeedbackInfo(ci, RedCorruptedContent, nsnull, 0);

Have an overload for connection entry instead of conn info?  We could have a static method to create the key from host,port,usingSSL,etc that nsHttpConnectionInfo can use and also this code can use.  Or, just find entry by nsIURI.

Creating nsHttpConnectionInfo for purpose of passing a hashing key doesn't seems right and necessary.

Both entries (anon/non-anon), if exist respectively, should get the feedback notification.

Comment 6

[Patrick McManus [:mcmanus]]

Attachment: patch v1

Comment 7

[Honza Bambas (:mayhemer)]

Comment on attachment 597440 [details] [diff] [review]
patch v1

Review of attachment 597440 [details] [diff] [review]:
-----------------------------------------------------------------

r=honzab

Comment 8

[Patrick McManus [:mcmanus]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/672cc4ee2dd9

Comment 9

[Matt Brubeck (:mbrubeck)]

Sorry, I backed this out on inbound because something in the push appeared to cause a leak in mochitest-browser-chrome:
https://hg.mozilla.org/integration/mozilla-inbound/rev/577da08878cb

Comment 10

[Patrick McManus [:mcmanus]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/6018a0d6b33d

Comment 11

[Marco Bonardo [:mak]]

https://hg.mozilla.org/mozilla-central/rev/6018a0d6b33d