Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Use px/agent_status_diagnostics script within px cli to detect missing kernel headers #2065

Conversation

ddelnano
Copy link
Member

@ddelnano ddelnano commented Dec 18, 2024

Summary: Use px/agent_status_diagnostics script within px cli to detect missing kernel headers

This PR leverages the script added in #2064 to detect missing kernel headers during cli deploys and px collect-logs commands. This solves 2/3 of the use cases I was hoping to identify for #2051 (the last being helm installs).

A recent example of this problem is #1986, where a Go TLS tracing bug went undiagnosed for months (August to December). Amazon Linux 2023's headers are different enough that it breaks Go TLS tracing when pixie's pre-packaged headers are used. The tooling in this PR would have provided a few opportunities for this to be caught.

Relevant Issues: #2051

Type of change: /kind feature

Test Plan: Verified the following scenarios

Test cases
  • px collect-logs works against a cloud that doesn't have a px/agent_status_diagnostics script
$ bazel run -c opt  --stamp src/pixie_cli:px -- collect-logs

WARN[0006] healthcheck script detected the following warnings:  error="Unable to detect if the cluster's nodes have the distro kernel headers installed (vizier too old to perform this check). Please ensure that the kernel headers are installed on all nodes."
Logs written to pixie_logs_20241223165214.zip

# zip file contains px/agent_status output
$ cat px_agent_diagnostics.txt
{"_tableName_":"output","agent_id":"07fb4d26-3b53-4ba7-9bb7-f2cb10a1e63d","asid":79,"hostname":"gke-dev-ddelnano1-default-pool-b099382d-30mu","ip_address":"","agent_state":"AGENT_STATE_HEALTHY","create_time":"2024-12-18T12:43:44.41952403Z","last_heartbeat_ns":4303060450,"kernel_headers_installed":true}
  • px collect-logs works against a cloud that does have a px/agent_status_diagnostics script
$ bazel run  src/pixie_cli:px -- collect-logs
INFO: Analyzed target //src/pixie_cli:px (0 packages loaded, 0 targets configured).
INFO: Found 1 target...
Target //src/pixie_cli:px up-to-date:
  bazel-bin/src/pixie_cli/px_/px
INFO: Elapsed time: 4.240s, Critical Path: 3.89s
INFO: 3 processes: 1 internal, 2 linux-sandbox.
INFO: Build completed successfully, 3 total actions
INFO: Running command line: bazel-bin/src/pixie_cli/px_/px collect-logs
Pixie CLI
*******************************
* ENV VARS
*        PX_CLOUD_ADDR=testing.getcosmic.ai:443
*******************************
Logs written to pixie_logs_20241218164734.zip

$ cat px_agent_diagnostics.txt
{"_tableName_":"output","headers_installed_percent":1}
  • px collect-logs identifies when kernel headers are missing when px/agent_status_diagnostics present
$ Logs written to pixie_logs_20241223165214.zip
$ bazel run -c opt  --stamp src/pixie_cli:px -- --bundle https://csmc-io.github.io/pxl-scripts/pxl_scripts/bundle.json collect-logs
[ ... ]
WARN[0012] healthcheck script detected the following warnings:  error="Detected missing kernel headers on your cluster's nodes. This may cause issues with the Pixie agent. Please install kernel headers on all nodes."

$ cat px_agent_diagnostics.txt
{"_tableName_":"output","headers_installed_percent":0.5}
  • Artificially forcing context deadline (timeout) results in an error
$ git diff
diff --git a/src/pixie_cli/pkg/vizier/script.go b/src/pixie_cli/pkg/vizier/script.go
index 7d3b7e008..c957b8943 100644
--- a/src/pixie_cli/pkg/vizier/script.go
+++ b/src/pixie_cli/pkg/vizier/script.go
@@ -317,7 +317,7 @@ func RunSimpleHealthCheckScript(br *script.BundleManager, cloudAddr string, clus
                execScript = br.MustGetScript(script.AgentStatusScript)
        }

-       ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+       ctx, cancel := context.WithTimeout(context.Background(), 1*time.Second)

$ bazel run  src/pixie_cli:px -- collect-logs

WARN[0012]src/pixie_cli/pkg/vizier/logs.go:135 px.dev/pixie/src/pixie_cli/pkg/vizier.(*LogCollector).CollectPixieLogs() failed to run health check script             error="context deadline exceeded"
Logs written to pixie_logs_20241218165033.zip
  • px collect-logs prompts auth flow when credentials don't match current cloud
$ PX_CLOUD_ADDR=new-cloud bazel run  src/pixie_cli:px -- collect-logs
*******************************
* ENV VARS
*        PX_CLOUD_ADDR=new-cloud
*******************************
Failed to authenticate. Please retry `px auth login`.
  • px deploy on pre v0.14.14 (older) vizier with existing bundle warns that kernel headers should be installed
# Additional flags provided to speed up vizier bootstrapping
$ bazel run -c opt --stamp src/pixie_cli:px -- deploy --pem_flags='PL_STIRLING_SOURCES=kNone' --deploy_key='<deploy key>' --deploy_olm=false --olm_namespace=olm --bundle=https://csmc-io.github.io/pxl-scripts/pxl_scripts/bundle.json
  • px deploy on pre v0.14.14 (older) vizier with latest bundle warns that kernel headers should be installed
# Additional flags provided to speed up vizier bootstrapping
$ bazel run -c opt --stamp src/pixie_cli:px -- deploy --pem_flags='PL_STIRLING_SOURCES=kNone' --deploy_key='<deploy key>' --deploy_olm=false --olm_namespace=olm --bundle=https://csmc-io.github.io/pxl-scripts/pxl_scripts/bundle.json

[ ... ]
Waiting for Pixie to pass healthcheck
 ✔    Wait for PEMs/Kelvin
 ✔    Wait for PEMs/Kelvin
 ✕    Wait for healthcheck  ERR: Unable to detect if the cluster's nodes have the distro kernel headers installed (vizier too old to perform this check). Please ensure that the kernel headers are installed on all nodes.
Pixie healthcheck detected the following warnings: error=Unable to detect if the cluster's nodes have the distro kernel headers installed (vizier too old to perform this check). Please ensure that the kernel headers are installed on all nodes.

[ ...]
  • px deploy on v0.14.14 vizier with latest bundle warns appropriate when kernel headers are missing
$ bazel run -c opt --stamp src/pixie_cli:px -- deploy --pem_flags='PL_STIRLING_SOURCES=kNone' --deploy_key=<deploy key> --bundle=https://csmc-io.github.io/pxl-scripts/pxl_scripts/bundle.json -v 0.14.14-pre-r1.0

[ ... ]
Waiting for Pixie to pass healthcheck
 ✔    Wait for PEMs/Kelvin
 ✕    Wait for healthcheck  ERR: Detected missing kernel headers on your cluster's nodes. This may cause issues with the Pixie agent. Please install kernel headers on all nodes.
Pixie healthcheck detected the following warnings: error=Detected missing kernel headers on your cluster's nodes. This may cause issues with the Pixie agent. Please install kernel headers on all nodes.

Changelog Message: Enhanced the px cli's deploy and collect-logs commands to surface when kernel headers aren't installed. This is a common source of bugs that can only be addressed by installing your distro's kernel headers.

@ddelnano ddelnano force-pushed the ddelnano/use-agent-diagnostics-px-deploy-collect-logs branch 4 times, most recently from 52994ea to d942389 Compare December 23, 2024 14:19
kernel headers

Signed-off-by: Dom Del Nano <ddelnano@gmail.com>
@ddelnano ddelnano force-pushed the ddelnano/use-agent-diagnostics-px-deploy-collect-logs branch from d942389 to b632cd4 Compare December 23, 2024 16:27
@ddelnano ddelnano marked this pull request as ready for review December 24, 2024 14:18
@ddelnano ddelnano requested a review from a team as a code owner December 24, 2024 14:18
@ddelnano
Copy link
Member Author

ddelnano commented Jan 2, 2025

@pixie-io/vizier please give this a look when you have the chance!

ddelnano added a commit that referenced this pull request Jan 6, 2025
Summary: Add px/agent_status_diagnostics pxl script

This PR adds a new pxl script that helps identify if linux headers are
missing on any PEMs. This can be extended in the future, but the first
use cause will be to execute the script during `px deploy` and `px
collect-logs`.

Relevant Issues: #2051

Type of change: /kind feature

Test Plan: Ran the script in the UI and tested it as part of the
validation for #2065

![Screen Shot 2024-12-18 at 10 51 15
AM](https://github.com/user-attachments/assets/bd4ab6de-ad92-4af3-8714-b044a7df7d65)


Changelog Message: Add `px/agent_status_diagnostics` pxl script for
checking common issues

Signed-off-by: Dom Del Nano <ddelnano@gmail.com>
@ddelnano ddelnano merged commit 3c9c4bd into pixie-io:main Jan 6, 2025
25 checks passed
@ddelnano ddelnano deleted the ddelnano/use-agent-diagnostics-px-deploy-collect-logs branch January 6, 2025 23:10
ddelnano added a commit to ddelnano/pixie that referenced this pull request Jan 16, 2025
ddelnano added a commit to ddelnano/pixie that referenced this pull request Jan 16, 2025
…ect missing kernel headers (pixie-io#2065)"

This reverts commit 3c9c4bd.

Signed-off-by: Dom Del Nano <ddelnano@gmail.com>
ddelnano added a commit to ddelnano/pixie that referenced this pull request Jan 16, 2025
…ect missing kernel headers (pixie-io#2065)"

This reverts commit 3c9c4bd.

Signed-off-by: Dom Del Nano <ddelnano@gmail.com>
aimichelle pushed a commit that referenced this pull request Jan 16, 2025
…ect missing kernel headers (#2065)" (#2081)

Summary: Revert "Use `px/agent_status_diagnostics` script within px cli
to detect missing kernel headers (#2065)"

This reverts commit 3c9c4bd. While
testing the latest cloud release, I noticed that the functionality added
in #2065 can cause `px deploy` to hang indefinitely at the "Wait for
healthcheck" step. The deploy will finish successfully, but it's a poor
user experience.

There seems to be a goroutine blocking issue that is dependent on
cluster size. On 1 and 2 node clusters that I tested #2065 on, the issue
doesn't surface. However, the issue reproduces reliably on the larger
clusters that have pixie deployed to. Let's revert and release a new cli
version once this is tracked down.

Relevant Issues: #2051

Type of change: /kind bugfix

Test Plan: N/A

Changelog Message: Reverted the recent advanced diagnostics added during
`px deploy` as in some cases it can cause that caused `px deploy` to
hang

Signed-off-by: Dom Del Nano <ddelnano@gmail.com>
ddelnano added a commit to ddelnano/pixie that referenced this pull request Jan 17, 2025
…ing kernel headers (pixie-io#2065)

Summary: Use `px/agent_status_diagnostics` script within px cli to
detect missing kernel headers

This PR leverages the script added in pixie-io#2064 to detect missing kernel
headers during cli deploys and `px collect-logs` commands. This solves
2/3 of the use cases I was hoping to identify for pixie-io#2051 (the last being
helm installs).

A recent example of this problem is
pixie-io#1986, where a Go TLS tracing
bug went undiagnosed for months (August to December). Amazon Linux
2023's headers are different enough that it breaks Go TLS tracing when
pixie's pre-packaged headers are used. The tooling in this PR would have
provided a few opportunities for this to be caught.

Relevant Issues: pixie-io#2051

Type of change: /kind feature

Test Plan: Verified the following scenarios
<details><summary>Test cases</summary>

- [x] `px collect-logs` works against a cloud that doesn't have a
`px/agent_status_diagnostics` script
```
$ bazel run -c opt  --stamp src/pixie_cli:px -- collect-logs

WARN[0006] healthcheck script detected the following warnings:  error="Unable to detect if the cluster's nodes have the distro kernel headers installed (vizier too old to perform this check). Please ensure that the kernel headers are installed on all nodes."
Logs written to pixie_logs_20241223165214.zip

# zip file contains px/agent_status output
$ cat px_agent_diagnostics.txt
{"_tableName_":"output","agent_id":"07fb4d26-3b53-4ba7-9bb7-f2cb10a1e63d","asid":79,"hostname":"gke-dev-ddelnano1-default-pool-b099382d-30mu","ip_address":"","agent_state":"AGENT_STATE_HEALTHY","create_time":"2024-12-18T12:43:44.41952403Z","last_heartbeat_ns":4303060450,"kernel_headers_installed":true}
```
- [x] `px collect-logs` works against a cloud that does have a
`px/agent_status_diagnostics` script
```
$ bazel run  src/pixie_cli:px -- collect-logs
INFO: Analyzed target //src/pixie_cli:px (0 packages loaded, 0 targets configured).
INFO: Found 1 target...
Target //src/pixie_cli:px up-to-date:
  bazel-bin/src/pixie_cli/px_/px
INFO: Elapsed time: 4.240s, Critical Path: 3.89s
INFO: 3 processes: 1 internal, 2 linux-sandbox.
INFO: Build completed successfully, 3 total actions
INFO: Running command line: bazel-bin/src/pixie_cli/px_/px collect-logs
Pixie CLI
*******************************
* ENV VARS
*        PX_CLOUD_ADDR=testing.getcosmic.ai:443
*******************************
Logs written to pixie_logs_20241218164734.zip

$ cat px_agent_diagnostics.txt
{"_tableName_":"output","headers_installed_percent":1}
```
- [x] `px collect-logs` identifies when kernel headers are missing when
`px/agent_status_diagnostics` present
```
$ Logs written to pixie_logs_20241223165214.zip
$ bazel run -c opt  --stamp src/pixie_cli:px -- --bundle https://csmc-io.github.io/pxl-scripts/pxl_scripts/bundle.json collect-logs
[ ... ]
WARN[0012] healthcheck script detected the following warnings:  error="Detected missing kernel headers on your cluster's nodes. This may cause issues with the Pixie agent. Please install kernel headers on all nodes."

$ cat px_agent_diagnostics.txt
{"_tableName_":"output","headers_installed_percent":0.5}
```

- [x] Artificially forcing context deadline (timeout) results in an
error
```
$ git diff
diff --git a/src/pixie_cli/pkg/vizier/script.go b/src/pixie_cli/pkg/vizier/script.go
index 7d3b7e008..c957b8943 100644
--- a/src/pixie_cli/pkg/vizier/script.go
+++ b/src/pixie_cli/pkg/vizier/script.go
@@ -317,7 +317,7 @@ func RunSimpleHealthCheckScript(br *script.BundleManager, cloudAddr string, clus
                execScript = br.MustGetScript(script.AgentStatusScript)
        }

-       ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+       ctx, cancel := context.WithTimeout(context.Background(), 1*time.Second)

$ bazel run  src/pixie_cli:px -- collect-logs

WARN[0012]src/pixie_cli/pkg/vizier/logs.go:135 px.dev/pixie/src/pixie_cli/pkg/vizier.(*LogCollector).CollectPixieLogs() failed to run health check script             error="context deadline exceeded"
Logs written to pixie_logs_20241218165033.zip
```
- [x] `px collect-logs` prompts auth flow when credentials don't match
current cloud
```
$ PX_CLOUD_ADDR=new-cloud bazel run  src/pixie_cli:px -- collect-logs
*******************************
* ENV VARS
*        PX_CLOUD_ADDR=new-cloud
*******************************
Failed to authenticate. Please retry `px auth login`.
```

- [x] `px deploy` on pre v0.14.14 (older) vizier with existing bundle
warns that kernel headers should be installed
```
# Additional flags provided to speed up vizier bootstrapping
$ bazel run -c opt --stamp src/pixie_cli:px -- deploy --pem_flags='PL_STIRLING_SOURCES=kNone' --deploy_key='<deploy key>' --deploy_olm=false --olm_namespace=olm --bundle=https://csmc-io.github.io/pxl-scripts/pxl_scripts/bundle.json
```

- [x] `px deploy` on pre v0.14.14 (older) vizier with latest bundle
warns that kernel headers should be installed
```
# Additional flags provided to speed up vizier bootstrapping
$ bazel run -c opt --stamp src/pixie_cli:px -- deploy --pem_flags='PL_STIRLING_SOURCES=kNone' --deploy_key='<deploy key>' --deploy_olm=false --olm_namespace=olm --bundle=https://csmc-io.github.io/pxl-scripts/pxl_scripts/bundle.json

[ ... ]
Waiting for Pixie to pass healthcheck
 ✔    Wait for PEMs/Kelvin
 ✔    Wait for PEMs/Kelvin
 ✕    Wait for healthcheck  ERR: Unable to detect if the cluster's nodes have the distro kernel headers installed (vizier too old to perform this check). Please ensure that the kernel headers are installed on all nodes.
Pixie healthcheck detected the following warnings: error=Unable to detect if the cluster's nodes have the distro kernel headers installed (vizier too old to perform this check). Please ensure that the kernel headers are installed on all nodes.

[ ...]
```

- [x] `px deploy` on v0.14.14 vizier with latest bundle warns
appropriate when kernel headers are missing
```
$ bazel run -c opt --stamp src/pixie_cli:px -- deploy --pem_flags='PL_STIRLING_SOURCES=kNone' --deploy_key=<deploy key> --bundle=https://csmc-io.github.io/pxl-scripts/pxl_scripts/bundle.json -v 0.14.14-pre-r1.0

[ ... ]
Waiting for Pixie to pass healthcheck
 ✔    Wait for PEMs/Kelvin
 ✕    Wait for healthcheck  ERR: Detected missing kernel headers on your cluster's nodes. This may cause issues with the Pixie agent. Please install kernel headers on all nodes.
Pixie healthcheck detected the following warnings: error=Detected missing kernel headers on your cluster's nodes. This may cause issues with the Pixie agent. Please install kernel headers on all nodes.
```

</details>

Changelog Message: Enhanced the `px` cli's `deploy` and `collect-logs`
commands to surface when kernel headers aren't installed. This is a
common source of bugs that can only be addressed by installing your
distro's kernel headers.

Signed-off-by: Dom Del Nano <ddelnano@gmail.com>
(cherry picked from commit 3c9c4bd)
aimichelle pushed a commit that referenced this pull request Jan 17, 2025
… missing kernel headers (#2091)

Summary: Re-introduce use of `px/agent_status_diagnostics` in px cli to
detect missing kernel headers

The original version of this was reverted since it resulted in hung `px
deploy` and `px collect-logs` commands on larger clusters. This PR
reintroduces the change with the fixes necessary to prevent the previous
issue and is best reviewed commit by commit as outlined below:

Commit 1: Cherry-pick of #2065
Commit 2: Fix for goroutine deadlock
Commit 3: Add bundle flag to the `px collect-logs` command
Commit 4: Introduce `PX_LOG_FILE` env var for redirecting `px` log
output to a file -- useful for debugging the cli since its terminal
spinners complicate logging to stdout

Commits 1, 3 and 4 should be self explanatory. As for Commit 2, the
goroutine deadlock occurred from the `streamCh` channel consumer.

The previous version read a single value from the `streamCh` channel,
parsed the result and
[terminated](2ec63c8#diff-4da8f48b4c664d330cff34e70f907d6015289797c832587b0b14004875ef0831R363)
its goroutine. Thus future sends to the `streamCh` channel could block
and prevent the pipe receiving the pxl script results to be fully
consumed. Since the stream adapter writes to the pipe, it couldn't flush
all of its results and the deadlock occurred.

The original testing was performed on clusters with 1 and 2 nodes -- max
of 2 PEMs and 2 results from `px/agent_status`. This deadlock issue
didn't surface in those situations because `streamCh` was a buffered
channel with capacity of 1 and the consumer would read a single record
before terminating. This meant that the pipe reader would hit EOF before
it would initiate a channel send that would deadlock as outlined below:

2 Node cluster situation:
1. `px` cli executes `px/agent_status` as `px/agent_status_diagnostics`
is not in the canonical bundle yet
2. streamCh producer sends 1st PEMs result -- streamCh at capacity
3. streamCh consumer reads the value and exits -- streamCh ready to
accept 1 value
4. streamCh producer sends 2nd and final PEM result -- streamCh at
capacity and future sends would block!
5. Program exits since pxl script is complete

Relevant Issues: #2051

Type of change: /kind feature

Test Plan: Verified that the deadlock no longer occurs on clusters with
3-6 nodes
- [x] Used the
[following](https://github.com/user-attachments/files/18457105/deadlocked-goroutines.txt)
pprof goroutine stack dump to understand the deadlock described above --
see blocked goroutine on `streamCh` channel send on `script.go:337`
- [x] Re-tested all of the scenarios from #2065

Changelog Message: Re-introduce enhanced diagnostics for `px deploy` and
`px collect-logs` commands used to detect common sources of environment
incompatibilities

---------

Signed-off-by: Dom Del Nano <ddelnano@gmail.com>
# for free to join this conversation on GitHub. Already have an account? # to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants