IIS collector might fail to initialize after reboot in v0.31.3

[JDA88]

Current Behavior

We deployed v0.31.3 on 450+ servers last week and we faced an odd behavior on the iis collector.
We had a few servers that failed to initialize the iis collector after a reboot. The collector continued to fail until we restarted the windows_exporter service.

On all the servers affected the behavior was the same:

• Both windows_exporter and W3SVC were on delayed start
• Restarting the windows_exporter service fixed the issue
• Event logs message where the same (details below)

Expected Behavior

Correct initialization of the iis connector on every reboot.
It was woking 100% in v0.29.1

Steps To Reproduce

Cannot be reproduced 100% of the time, but:
- Start windows_exporter before the W3SVC is 100% initialized

Environment

• windows_exporter Version: 0.31.3
• Windows Server Version: 2019/2022/2025

windows_exporter logs

At startup:

source=collection.go:248 msg="couldn't initialize collector" err="error build collector iis: failed to build APP_POOL_WAS collector: failed to create APP_POOL_WAS collector: failed to initialize collector: GetCounterTimeBase: No data to return.\r\n\nGetCounterTimeBase: No data to return.\r\n\nGetCounterTimeBase: No data to return.\r\n"

Then after, on every call:
source=collect.go:220 msg="collector iis failed after 9.1212ms, resulting in 99 metrics" err="failed to collect APP_POOL_WAS metrics: failed to collect APP_POOL_WAS metrics: performance counter not initialized. Check application logs from initialization pharse for more information"

Anything else?

Counters are 100% fine on all those servers as restarting the service fixe the issues, no other messages in event logs, iis working fine also.
We had the issue on 2 servers on 22 where the iis collector is enabled, too soon to judge but looks like a 10% average failure rate after reboot. We do pause the v0.31.3 version deployment waiting for more information’s.

We do use the --collector.iis.app-include option to filter instance but I do not think this can make a difference.

[jkroepke]

Hm, currently the windows_exporter initialize all collector at startup time. if the PDH interface mention that there is not data.. not sure how I could recover that.

[JDA88]

I understand. We delayed the start of the service and the occurrence have dropped.

What bothered me was the fact that the with the previous version we deployed never experienced this.

For us the issue is not blocking anymore. You can close if you find the behavior "as expected"

PS: I'm not sure if I understood correctly but maybe adding a "retry to initialize every min for x min if uptime bellow 5-10 min" on some collectors that might experience this could help?