Gerrit server displays "Server unavailble" (Server is up)

[Ray Pelkey]

on Gerrit 2.11.7 some times users will see a Server Unavailable message (See attachment) but a retry with the browser immediately after logs them in successfully and the server is running during these events.

any advice on what could be causing this?  Network problems so far have been ruled out.

[Ray Pelkey]

btw, the maxThreads in http is set to 30

[Ray Pelkey]

On Friday, May 13, 2016 at 6:30:52 AM UTC-4, Ray Pelkey wrote:

btw, the maxThreads in httpd section is set to 30.

[Gaurav Negi]

Hi Ray, we are also using Gerrit 2.11.7 and we are not seen any such issue. 

Any particular this issue is shown? On Chrome we are not hitting this.

Sent from my iPhone

--
--
To unsubscribe, email repo-discuss...@googlegroups.com
More info at http://groups.google.com/group/repo-discuss?hl=en

---
You received this message because you are subscribed to the Google Groups "Repo and Gerrit Discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to repo-discuss...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ray Pelkey]

Hi Gaurav,

Thanks for the reply! 

In this particular case, only one user has mentioned this to us. We have around 2,000 users currently who are connecting from all over the globe.  It's a little peculiar that more people are asking about this.

The browser in use at the time was Firefox, but chrome was tried recently, and it seemed to not be a problem but we've not heard any updates for a few hours now.

[Ray Pelkey]

typo,  that is to say:

  It's a little peculiar that more people are NOT asking about this. ...

sorry about that,

[Martin Fick]

On Friday, May 13, 2016 03:23:09 AM Ray Pelkey wrote:
> on Gerrit 2.11.7 some times users will see a Server
> Unavailable message (See attachment) but a retry with the
> browser immediately after logs them in successfully and
> the server is running during these events.

You say "logs them in", does that mean you believe they were
not logged in (maybe their sessions expired), when the
original failures happened? Does this happen mostly on
Monday mornings?

> any advice on what could be causing this? Network
> problems so far have been ruled out.

Can you post all of your http configs? Also, can you
describe your http setup, are you using apache or another
web server/proxy in front of Gerrit?

Do you have any reason to believe this is not a load
problem, too many users connecting at once?

-Martin

--
The Qualcomm Innovation Center, Inc. is a member of Code
Aurora Forum, hosted by The Linux Foundation

[Ray Pelkey]

Hi Martin

<You say "logs them in", does that mean you believe they were
<not logged in (maybe their sessions expired), when the
<original failures happened?  Does this happen mostly on
<Monday mornings?

I think Sebastian did mention that it's more common earlier in the week.

He can see the gerrit ui in the background, with the server unavailable pop up
(attached screen cap)

what's odd, only one user seeing this at this time, no one else has mentioned this.

< Can you post all of your http configs?

no proxy, just httpd (appache) server

http related sections of our config:
[httpd]
        listenUrl = proxy-https://*:80/
        requestLog = true
        maxThreads = 30

[cache "web_sessions"]
       maxAge = 1 w

<Do you have any reason to believe this is not a load
<problem, too many users connecting at once?

Possible, httpd.maxThreads set at 30 may be too low for us, we have quite a few users, any other httpd setting is just default.

another browser was going to be checked out, currently using firefox, we are waiting to see if there's any difference with chrome.

[Martin Fick]

On Monday, May 16, 2016 03:17:30 AM Ray Pelkey wrote:
> <You say "logs them in", does that mean you believe they
> were <not logged in (maybe their sessions expired), when
> the <original failures happened? Does this happen mostly
> on <Monday mornings?
>
> I think Sebastian did mention that it's more common
> earlier in the week.

People's sessions often expire over the weekend. This can
cause several issues.

1) It means that people who have browser windows that are
open are poting to changes, may suddenly fail RPC calls.

2) It also means that load usually goes up on Monday
mornings because of all the changes that now are no longer
visible to most users since they are no longer logged in.
Scanning the open changes pages becomes extremely slow since
most changes are rejected (not logged in, and thus not
visible) and the scan must go much further, sometimes
scanning all open changes. You can usually see this load
increase on your db,