New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 891940 link

Starred by 2 users

Issue metadata

Status: Fixed
Owner:
Closed: Oct 4
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 3
Type: Bug



Sign in to add a comment

Network requests stuck in pending state

Project Member Reported by blois@google.com, Oct 3

Issue description

Chrome Version       : Chrome 69
URLs (if applicable) : https://colab.corp.google.com
Other browsers tested:
     Safari:
    Firefox:
       Edge:

What steps will reproduce the problem?
(1) File-> Save a copy in Drive.
(2) In the new document, save multiple times.
(3)

What is the expected result?
All saves complete.


What happens instead?
A number of our users are seeing the save operations hang.

We've had difficulty reproducing this issue ourselves, but a significant number of Googlers have been reporting this. Reports started in early August and have been ramping up since then.

See:
 b/112431524
 b/116274455
 b/117235307


From b/117235307 we were able to capture a trace with chrome://net-export/ as the user was experiencing the issue and near the end of the trace we see:
t=7743410 [st=    0] +REQUEST_ALIVE  [dt=25531+]
                      --> priority = "LOWEST"
                      --> url = "https://clients6.google.com/upload/drive/v2beta/files..."
t=7743410 [st=    0]   +DELEGATE_INFO  [dt=25531+]
                        --> delegate_blocked_by = "ResourceScheduler"
t=7768941 [st=25531] 

The trace continues for another 20 seconds but there is no progress on these requests. Access to the trace can be requested from https://drive.google.com/file/d/1UO5Q1lz0y7zz0xfeBf5nyltGyHEP7RiA/view

Also from b/117235307 you'll see that the network requests are stuck in pending state.

Similarly, https://b.corp.google.com/issues/112431524#comment23 has a number of network requests stuck in pending state.
 
Cc: piatek@google.com jasnyder@google.com craigcitro@google.com clouser@google.com vanderplas@google.com fischman@google.com
Note that we've only had reports from @google.com users, no external users have reported it on the external Colaboratory. One difference between the two is that external Colaboratory has 2 fewer long-polling network requests continually running.

Not sure if this is related to crbug.com/837771.

Labels: Needs-Triage-M69
Cc: yhirano@chromium.org
Owner: tbansal@chromium.org
Seems blocked in the chrome network stack rather than blink.
//services/network/resource_scheduler.cc

This also may be related to the socket pool issue of 872284

tbansal@, IIRC, you ran an experiment in this layer. Could you take a look or find a right person to investigate the problem?
Status: Started (was: Unconfirmed)
I can confirm that it can happen when the user is in 3 experiments (all of which I am running).  The bug does not affect external users, and affects ~15% of Google users. To repro this, start Chrome with switch:  --force-effective-connection-type=3G. The experiments kick in only when the user is signed in with google.com account in Chrome. 

The 3 experiments:

1. DelayRequestsOnMultiplexedConnections: Enabling this experiment allows Chrome to throttle requests to HTTP 2.0 servers.

2. LowPriorityIframes: This causes requests in iframes to be marked as low priority, and can cause them to be throttled in presence of high priority requests.

3. ThrottleDelayable: This causes Chrome to start throttling low priority requests if there are too many high priority requests in-flight.

When all 3 experiments are enabled, and when effective-connection-type is detected as 3G, then XHRss from iframes will be marked as low priority, and so can be throttled if there are too many high priority requests in-flight (in this case: XHRs from main frame).

I am going to disable experiment (2) and (3) for now which we were not planning to launch anyways.
Status: Fixed (was: Started)
This should be fixed now (pending the user restarting Chrome).
Hey Tarun, let me confirm that this issue won't happen after m70, right?
Or still need to wait a target mileston?
Labels: M-69
This was a Finch experiment, and has been turned off. Users should not have to wait for this to be fixed. M-69 label seems appropriate.
I see, perfect!

Sign in to add a comment