BOSH Director Time Drift


Earlier today I was greeted with this message while performing a BOSH deployment for SHIELD:

Error 100: Unknown CPI error 'Unknown' with message 'AWS was not able to validate the provided access credentials'

I double checked the AWS Access and Secret permissions. I even deployed from my laptop a test deployment with the same keys with no issue. So the problem was not with the keys.

After digging around I found this article which pointed me to check the time on the BOSH Director. Why? The BOSH AWS CPI is using the AWS CLI to perform actions which leverage a token which is time sensitive.

To verify compare the time at Google to the local server:

root:~# date
Wed Jan 18 15:30:50 UTC 2017
[email protected]:~# sudo date -s "$(wget -qSO- --max-redirect=0 2>&1 | grep Date: | cut -d' ' -f5-8)Z"
Wed Jan 18 15:27:50 UTC 2017

The BOSH Director was living 3 minutes in the future. A quick resetting of the clock:

sudo date --set="2017-01-18 15:30:00.000"

Running bosh deploy again ran successfully after the change.

The long term solution is to figure out why NTP didn’t keep the date in sync. By default there is a cron job for root which is supposed to update ntp every 15 minutes:

root:~ crontab -l
0,15,30,45 * * * * /var/vcap/bosh/bin/ntpdate
0 0 * * * /var/vcap/jobs/director/bin/task_logrotate >>/var/vcap/sys/log/director/task_logrotate.log 2>&1

I’ll update this post once I figure out what the issue is with the two defined NTP servers.

Spread the word

twitter icon facebook icon linkedin icon