ADU Agent has very high CPU Usage when disconnected

[agreentree-dmc]

I am running the ADU Linux Agent and it is configured to use Azure Identity Service for connection information.

Before the Azure Identity Service has been configured, or if it's been configured incorrectly. The Azure Device Update linux agent periodically takes up 100% CPU usage for 3-4 seconds. I've provided screenshots for the CPU usage and for the Device Update Agent logs when this occurs.

I also see this behavior if the DeviceUpdate Agent has connected before, but is currently unable to connect (example: the device is running with no internet connection)

This behavior is undesirable. Is there any configuration I can setup to avoid this or any agent code changes I could make?

[jw-msft]

For repro, could you provide an strace and 2 ltrace (one with call summary and one without)?

systemctl stop deviceupdate-agent.service

STrace

strace -o ~/strace.log -f -tt -yy /usr/bin/AducIotAgent -l0

LTrace (No call summary)

# Note: no lowercase -c
ltrace -o ~/ltrace_no_summary.log -C -f /usr/bin/AducIotAgent

LTrace (with summary)

ltrace -o ~/ltrace.log -c -C -f /usr/bin/AducIotAgent

[agreentree-dmc]

@jw-msft Would you like the test to be run:

• When the device has not yet connected for the first time?
• When the device is fully connected?
• When the device has connected before, but now is not connection (ex. no network access on device)?

Additionally, I can provide the strace but my system does not have the ltrace command - is there an alternative test or command that I could try?

[agreentree-dmc]

@jw-msft Can you please provide more details on what you are looking for so that I can provide?

[jw-msft]

I can't know for sure, but it may be related to the process restarting itself via the exit in this scenario ( and systemd is setup to launch it if not running) in essentially a tight loop. A mitigation fix might be to have the agent process sleep for a second or so when first starting up or to sleep a while before exiting.

However, as a workaround, would it be possible to ensure that AIS is configured before starting ADU Agent or to ensure that AIS is configured correctly? It was mentioned that those were the cases where the high cpu usage behavior was observed, so avoiding those situations could likely work around this case.

As for what scenario the strace should be run, any repro scenario would do. However, it may not be helpful if what's causing it is a tight loop of process restarts. I'm mostly interested in if there is something else going on from a syscall/library call perspective within the run of a single process PID instance.

[agreentree-dmc]

@jw-msft
Thanks for your feedback. I'm pretty confident the agent restarting in a tight loop is not the problem, because we've had the RestartSec parameter on the service which is starting the agent set to 60 seconds even when observing the problem.

Yes - as the work around not we added a check that confirms AIS has made a successful connection before starting up.

Regarding the strace - I will have my colleague follow up with the results.