No System logs are generated for Linux agent, when agent is installed with --unprivileged flag.

[amolnater-qasource]

Kibana Build details:

VERSION: 8.13.0-SNAPSHOT
BUILD: 70749
COMMIT: a0f4897f7c04069faf2a86dbda1dabea78c161c1
Artifact Link: https://snapshots.elastic.co/8.13.0-l534sdis/downloads/beats/elastic-agent/elastic-agent-8.13.0-SNAPSHOT-linux-x86_64.tar.gz

Host OS: Linux- SLES15, Ubuntu 22

Preconditions:

1. 8.13.0 Snapshot linux agent should be installed.
2. Linux Agents should be installed using below command:
   sudo ./elastic-agent install --url=<url> --enrollment-token=<token> --unprivileged

Steps to reproduce:

1. Navigate to Data Streams tab.
2. Select Type: logs, System Integration, and required namespace filters.
3. Observe no System integration logs for linux agent installed with --unprivileged flag.

What's working fine:

• Issue is not reproducible on Linux- SLES15, Ubuntu 22 when agents are installed without --unprivileged flag.

Screen Recording:

Data.streams.-.Fleet.-.Elastic.-.Google.Chrome.2024-01-22.13-59-51.mp4

Logs:
elastic-agent-diagnostics-2024-01-22T08-42-48Z-00.zip
elastic-agent-diagnostics-2024-01-22T08-42-55Z-00.zip

Expected Result:
System logs should be generated for Linux agent, when agent is installed with --unprivileged flag.

Feature:
https://github.com/elastic/ingest-dev/issues/1766

[elasticmachine]

Pinging @elastic/elastic-agent-control-plane (Team:Elastic-Agent-Control-Plane)

[amolnater-qasource]

@manishgupta-qasource Please review.

[manishgupta-qasource]

Secondary review for this ticket is Done

[pierrehilbert]

@blakerouse could you please have a look here?
I can see some errors:
{"log.level":"debug","@timestamp":"2024-01-22T08:40:56.617Z","message":"Error fetching PID info for 1, skipping: FillPidMetrics: error fetching IO metrics for pid 1: error fetching IO metrics: open /proc/1/io: permission denied","component":{"binary":"metricbeat","dataset":"elastic_agent.metricbeat","id":"system/metrics-default","type":"system/metrics"},"log":{"source":"system/metrics-default"},"log.origin":{"file.line":173,"file.name":"process/process.go","function":"github.com/elastic/elastic-agent-system-metrics/metric/system/process.(*Stats).pidIter"},"service.name":"metricbeat","ecs.version":"1.6.0","log.logger":"processes","ecs.version":"1.6.0"}

[cmacknz]

This is not a bug, this is expected. We are likely also missing several metrics, that error looks like we can't read /proc/1 which would be the process information for the PID 1 or the init system. The system integration is probably going to have the most problems operating as non-root.

Copying a previous comment on this:

In the case of the system integration, on Linux we include reading the contents of /var/log/auth.log and /var/log/syslog by default but non-root users cannot read these on recent Ubuntu versions. The auth and system log datastreams are going to be empty.

testuser@valuable-gudgeon:/home/ubuntu$ lsb_release -a
No LSB modules are available.
Distributor ID:	Ubuntu
Description:	Ubuntu 22.04.3 LTS
Release:	22.04
Codename:	jammy
testuser@valuable-gudgeon:/home/ubuntu$ ls -la /var/log/
total 1452
drwxrwxr-x   8 root      syslog             4096 Jan 11 14:34 .
drwxr-xr-x  13 root      root               4096 Dec 10 21:18 ..
-rw-r--r--   1 root      root                452 Dec 10 21:19 alternatives.log
drwxr-xr-x   2 root      root               4096 Dec 10 21:19 apt
-rw-r-----   1 syslog    adm               11988 Jan 11 14:41 auth.log
-rw-rw----   1 root      utmp                  0 Dec 10 21:17 btmp
-rw-r-----   1 root      adm                7516 Jan 11 14:34 cloud-init-output.log
-rw-r-----   1 syslog    adm              204731 Jan 11 14:34 cloud-init.log
drwxr-xr-x   2 root      root               4096 Aug  2 11:53 dist-upgrade
-rw-r-----   1 root      adm               35909 Jan 11 14:34 dmesg
-rw-r-----   1 root      adm               37307 Jan  9 15:27 dmesg.0
-rw-r--r--   1 root      root               7904 Dec 10 21:19 dpkg.log
drwxr-sr-x+  3 root      systemd-journal    4096 Jan  9 15:26 journal
-rw-r-----   1 syslog    adm               92532 Jan 11 14:39 kern.log
drwxr-xr-x   2 landscape landscape          4096 Jan  9 15:26 landscape
-rw-rw-r--   1 root      utmp             296592 Jan 11 14:40 lastlog
drwx------   2 root      root               4096 Jan  9 15:26 private
-rw-r-----   1 syslog    adm             1030960 Jan 11 14:47 syslog
drwxr-x---   2 root      adm                4096 Jan  9 15:26 unattended-upgrades
-rw-rw-r--   1 root      utmp               6400 Jan 11 14:34 wtmp
testuser@valuable-gudgeon:/home/ubuntu$ cat /var/log/syslog
cat: /var/log/syslog: Permission denied
testuser@valuable-gudgeon:/home/ubuntu$ cat /var/log/auth.log
cat: /var/log/auth.log: Permission denied

In this case the work around would be to add the elastic-agent user to the adm group as that is the group with access to read those log files. Something similar may work /proc.

We need to decide if this needs to be an action taken by the agent or if it has to be a manual step from the user.

[pierrehilbert]

This is what we mentioned in the past: testing will allow us to highlight what is not working as expected and we will have to decide what solution we will pick for those failing:

• Fix the issue (by giving more permissions to the user, changing the behavior, etc.)
• Degrade the integration (for example only getting processes you are allowed to see and document how to get more)
• Make it clear that this integration is not supported when Agent is run unprivileged

[cmacknz]

We need to go through the system integration datastream by datastream on each OS and compare what it produces with --unpriviledged to what it produces without it. There are more issues than just this one lurking here.

[blakerouse]

@cmacknz is correct, this is expected. I do wonder if metrics are collected for the PID's that it can read, I would hope that metricbeat just continues on and reads the metrics for the PID's it can.

[nimarezainia]

I don't think agent should do anything to change group ownership. This is an admin decision. I see it our job to educate them to do this - if this datastream is needed by them.

In order to do that what we need is:

• Highlight the issue via degradation of the integration
• Enhance the documentation to help the user navigate what they need to do (with a tool-tip or pointer from Fleet UI)

if you agree can we use this issue for the first point?

[nimarezainia]

@amolnater-qasource thanks for this. Is this the only issue you see with System Integration? could we get the full list of issues for tracking purposes. thanks

[amolnater-qasource]

Hi @nimarezainia

The test plan is in progress and we haven't encountered issues for any other integrations as of now.

• System Metrics ✅
• Linux Integration ✅
• Custom Logs ✅
• Nginx ✅
• Redis ✅
• MySQL ✅
• Kafka ✅
• RabbitMQ ✅
• Osquery Manager ✅
• Elastic Defend ❌

Further, Elastic Defend is not supported as of now so the related feature testcases are also BLOCKED.

Screenshot:

We are executing test plan at link Fleet Feature Regression Test plan

Please let us know if we are missing anything here.
Thanks!

[jlind23]

Highlight the issue via degradation of the integration

@nimarezainia shall we just highlight that it is not going to fully work or shall we details what is the exact data that will not be collected?

[jlind23]

Closing this as covered by elastic/beats#39733