After an appliance reboot, the UI login spinner spins forever

[jrafanie]

This might be related to #3444

The UI worker strace is showing futex(0xca9634, FUTEX_WAIT_PRIVATE, 1721, NULL)

Note, I'm able to wget localhost:3000 so it is responding to requests and works but the UI worker seems stuck via strace.

To recreate this: you need to ensure 7a05adb is applied.
The initial deployment of the appliance seems to work (logging in and navigating around works). After you reboot the appliance, that's when you get the login page but it spins forever.

Reverting that commit and setting the database.yml pool to 1, bundling and things work again.

[chessbyte]

@dmetzger57 Weren't you looking at this today?

[dmetzger57]

Yes, LJ kindly helped me triage and we discovered the issue is related to our move to Puma.

On Jul 15, 2015, at 12:55 PM, Oleg Barenboim [email protected] wrote:

@dmetzger57 https://github.com/dmetzger57 Weren't you looking at this today?

—
Reply to this email directly or view it on GitHub #3468 (comment).

[Fryguy]

Note that @hsong-rh found you can still type through the spinner...so you can hit enter and login anyway.

[imtayadeway]

FWIW i can still reproduce the bug with the connection pool set to 1 (but still running puma). restarting doesn't seem to be necessary - just logging out/in again. if i hit refresh, sometimes i am logged in and sometimes i am taken back to the login page, where i can then log in successfully.

some other odd behavior that @gtanzillo found that i have also reproduced: hitting refresh will sometimes result in some HTML being printed rather than rendered.

[imtayadeway]

Found some things of interest when comparing a happy request to POST dashboard/authenticate (top in each screenshot) with an unhappy one (bottom in each). Some things to note:

1. The response time of an unhappy request (20+ seconds)
2. The difference in response body
3. The XML parsing error in the unhappy request

[matthewd]

I believe this is puma/puma#743

[matthewd]

I've updated the Gemfile on the appliance to pull puma from git, and done an evm:restart.

I wasn't doing a very good job of reproducing it, however, so... let me know whether you can still get it to break? ¯_(ツ)_/¯