usm: watcher: Add periodic scan #32400
Conversation
There are cases where the eBPF shared library watcher can miss hooking shared libraries in running processes. For example: (1) When a process such as nginx daemonizes itself by forking. Depending on timing, the parent process may exit before the hooking is complete leading to the hooking of the library failing, and the child process will never be noticied by the watcher since it never opens the shared library (since it's inherited from the parent). (2) When the path length is longer than what the eBPF program supports. To ensure that we don't miss these kind of cases, introduce a periodic scan to try to attach to running processes. We maintain a map to ensure that we don't scan processes over and over again to limit CPU usage. We scan every processes twice in order to avoid races with startup where we may scan a process' maps file before it has opened its shared libraries.
github-actions
bot
added
component/system-probe
medium review
vitkyrka
added
changelog/no-changelog
team/usm
|
[Fast Unit Tests Report] On pipeline 51562183 (CI Visibility). The following jobs did not run any unit tests: Jobs:
If you modified Go files and expected unit tests to run in these jobs, please double check the job logs. If you think tests should have been executed reach out to #agent-devx-help |
Test changes on VMUse this command from test-infra-definitions to manually test this PR changes on a VM: inv aws.create-vm --pipeline-id=51562183 --os-family=ubuntuNote: This applies to commit db4451f |
Uncompressed package size comparisonComparison with ancestor Diff per package
Decision✅ Passed |
Regression DetectorRegression Detector ResultsMetrics dashboard Baseline: b5ed471 ❌ Experiments with missing or malformed dataThis is a critical error. No usable optimization goal data was produced by the listed experiments. This may be a result of misconfiguration. Ping #single-machine-performance and we can help out.
Optimization Goals: ✅ No significant changes detected
|
| perf | experiment | goal | Δ mean % | Δ mean % CI | trials | links |
|---|---|---|---|---|---|---|
| ➖ | otel_to_otel_logs | ingress throughput | +0.09 | [-0.55, +0.73] | 1 | Logs |
| ➖ | quality_gate_idle | memory utilization | +0.04 | [+0.00, +0.08] | 1 | Logs bounds checks dashboard |
| ➖ | file_to_blackhole_300ms_latency | egress throughput | +0.02 | [-0.63, +0.67] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency_http2 | egress throughput | +0.01 | [-0.87, +0.89] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency | egress throughput | +0.01 | [-0.87, +0.89] | 1 | Logs |
| ➖ | tcp_dd_logs_filter_exclude | ingress throughput | -0.00 | [-0.01, +0.01] | 1 | Logs |
| ➖ | file_tree | memory utilization | -0.01 | [-0.14, +0.12] | 1 | Logs |
| ➖ | file_to_blackhole_100ms_latency | egress throughput | -0.01 | [-0.69, +0.67] | 1 | Logs |
| ➖ | file_to_blackhole_1000ms_latency | egress throughput | -0.05 | [-0.84, +0.75] | 1 | Logs |
| ➖ | file_to_blackhole_500ms_latency | egress throughput | -0.06 | [-0.84, +0.72] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency_http1 | egress throughput | -0.07 | [-0.98, +0.84] | 1 | Logs |
| ➖ | file_to_blackhole_1000ms_latency_linear_load | egress throughput | -0.10 | [-0.56, +0.37] | 1 | Logs |
| ➖ | quality_gate_idle_all_features | memory utilization | -0.10 | [-0.18, -0.02] | 1 | Logs bounds checks dashboard |
| ➖ | uds_dogstatsd_to_api_cpu | % cpu utilization | -0.24 | [-0.93, +0.45] | 1 | Logs |
| ➖ | tcp_syslog_to_blackhole | ingress throughput | -0.28 | [-0.37, -0.18] | 1 | Logs |
| ➖ | quality_gate_logs | % cpu utilization | -0.50 | [-3.73, +2.73] | 1 | Logs |
Bounds Checks: ❌ Failed
| perf | experiment | bounds_check_name | replicates_passed | links |
|---|---|---|---|---|
| ❌ | quality_gate_idle | memory_usage | 9/10 | bounds checks dashboard |
| ❌ | quality_gate_logs | memory_usage | 9/10 | |
| ✅ | file_to_blackhole_0ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http1 | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http1 | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http2 | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http2 | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_1000ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_1000ms_latency_linear_load | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_100ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_100ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_300ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_300ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_500ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_500ms_latency | memory_usage | 10/10 | |
| ✅ | quality_gate_idle_all_features | memory_usage | 10/10 | bounds checks dashboard |
| ✅ | quality_gate_logs | lost_bytes | 10/10 |
Explanation
Confidence level: 90.00%
Effect size tolerance: |Δ mean %| ≥ 5.00%
Performance changes are noted in the perf column of each table:
- ✅ = significantly better comparison variant performance
- ❌ = significantly worse comparison variant performance
- ➖ = no significant change in performance
A regression test is an A/B test of target performance in a repeatable rig, where "performance" is measured as "comparison variant minus baseline variant" for an optimization goal (e.g., ingress throughput). Due to intrinsic variability in measuring that goal, we can only estimate its mean value for each experiment; we report uncertainty in that value as a 90.00% confidence interval denoted "Δ mean % CI".
For each experiment, we decide whether a change in performance is a "regression" -- a change worth investigating further -- if all of the following criteria are true:
-
Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.
-
Its 90.00% confidence interval "Δ mean % CI" does not contain zero, indicating that if our statistical model is accurate, there is at least a 90.00% chance there is a difference in performance between baseline and comparison variants.
-
Its configuration does not mark it "erratic".
CI Pass/Fail Decision
❌ Failed. Some Quality Gates were violated.
- quality_gate_idle_all_features, bounds check memory_usage: 10/10 replicas passed. Gate passed.
- quality_gate_logs, bounds check lost_bytes: 10/10 replicas passed. Gate passed.
- quality_gate_logs, bounds check memory_usage: 9/10 replicas passed. Failed 1 which is > 0. Gate FAILED.
- quality_gate_idle, bounds check memory_usage: 9/10 replicas passed. Failed 1 which is > 0. Gate FAILED.
| @@ -26,7 +26,7 @@ import ( | |||
| "github.com/DataDog/datadog-agent/pkg/util/log" | |||
| ) | |||
|
|
|||
| const ( | |||
| var ( | |||
There was a problem hiding this comment.
The test changes this to a much smaller value in order to avoid waiting for a long time.
There was a problem hiding this comment.
Can you leave a comment in the code then?
| // findDeletedProcesses returns the terminated PIDs from the given map. | ||
| func findDeletedProcesses[V any](pids map[uint32]V) map[uint32]struct{} { | ||
| existingPids := make(map[uint32]struct{}, len(pids)) | ||
| func (w *Watcher) sync() { |
There was a problem hiding this comment.
Can you add a documentation for this function?
| // libraries. Scanning twice with the sync interval reduce this risk of | ||
| // missing shared libraries due to this. | ||
| if scanned == 0 || scanned == 1 { | ||
| w.scannedPIDs[pid] = scanned + 1 |
There was a problem hiding this comment.
| w.scannedPIDs[pid] = scanned + 1 | |
| w.scannedPIDs[pid]++ |
| // just after it has been exec'd and before it has opened its shared | ||
| // libraries. Scanning twice with the sync interval reduce this risk of | ||
| // missing shared libraries due to this. | ||
| if scanned == 0 || scanned == 1 { |
There was a problem hiding this comment.
| if scanned == 0 || scanned == 1 { | |
| if scanned < 2 { |
What does this PR do?
There are cases where the eBPF shared library watcher can miss hooking shared libraries in running processes. For example:
(1) When a process such as nginx daemonizes itself by forking.
Depending on timing, the parent process may exit before the hooking is
complete leading to the hooking of the library failing, and the child
process will never be noticied by the watcher since it never opens the
shared library (since it's inherited from the parent).
(2) When the path length is longer than what the eBPF program supports.
To ensure that we don't miss these kind of cases, introduce a periodic scan to try to attach to running processes. We maintain a map to ensure that we don't scan processes over and over again to limit CPU usage. We scan every processes twice in order to avoid races with startup where we may scan a process' maps file before it has opened its shared libraries.
Motivation
https://datadoghq.atlassian.net/wiki/spaces/UT/pages/4105764896
Describe how you validated your changes
Automated test added.
Also verified manually that with the fix
nginxis monitored if launched like thisBy checking the debug endpoint
sudo curl --unix /opt/datadog-agent/run/sysprobe.sock http://unix/network_tracer/debug/usm/traced_programs | jq | gronPossible Drawbacks / Trade-offs
Additional Notes