[Dynamic Instrumentation] Fix stability issues

[GreenMatan]

What does this PR do?

• Fixed few nil dereferences that caused crashes (DEBUG-3454)
• Fixed a race condition between map reading & writing, resulting in a crash (DEBUG-3211)
• One probe used to sabotage all other probes in a service if there was a failure during DWARF inspection (DEBUG-3230)
• Similar to the previous point, one process used to fail all other processes inspection if there was an error while processing probes for particular process (DEBUG-3205)

Motivation

Improve stability of Go DI

Describe how you validated your changes

I ran locally our e2e tests & exploration tests and verified the proposed changes improved our stability.

[bits-bot]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[agent-platform-auto-pr]

[Fast Unit Tests Report]

On pipeline 56883287 (CI Visibility). The following jobs did not run any unit tests:

Jobs:

• tests_windows-x64

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

[agent-platform-auto-pr]

Uncompressed package size comparison

Comparison with ancestor 153cd143979ee9acb303aa22608e636442089005

Diff per package

package	diff	status	size	ancestor	threshold
datadog-agent-amd64-deb	0.00MB	✅	835.77MB	835.76MB	0.50MB
datadog-agent-x86_64-rpm	0.00MB	✅	845.56MB	845.55MB	0.50MB
datadog-agent-x86_64-suse	0.00MB	✅	845.56MB	845.55MB	0.50MB
datadog-agent-arm64-deb	0.00MB	✅	826.28MB	826.28MB	0.50MB
datadog-agent-aarch64-rpm	0.00MB	✅	836.05MB	836.05MB	0.50MB
datadog-dogstatsd-amd64-deb	0.00MB	✅	39.42MB	39.42MB	0.50MB
datadog-dogstatsd-x86_64-rpm	0.00MB	✅	39.50MB	39.50MB	0.50MB
datadog-dogstatsd-x86_64-suse	0.00MB	✅	39.50MB	39.50MB	0.50MB
datadog-dogstatsd-arm64-deb	0.00MB	✅	37.96MB	37.96MB	0.50MB
datadog-heroku-agent-amd64-deb	0.00MB	✅	443.28MB	443.28MB	0.50MB
datadog-iot-agent-amd64-deb	0.00MB	✅	62.02MB	62.02MB	0.50MB
datadog-iot-agent-x86_64-rpm	0.00MB	✅	62.09MB	62.09MB	0.50MB
datadog-iot-agent-x86_64-suse	0.00MB	✅	62.09MB	62.09MB	0.50MB
datadog-iot-agent-arm64-deb	0.00MB	✅	59.27MB	59.27MB	0.50MB
datadog-iot-agent-aarch64-rpm	0.00MB	✅	59.33MB	59.33MB	0.50MB

Decision

✅ Passed

[cit-pr-commenter]

Regression Detector

Regression Detector Results

Metrics dashboard
Target profiles
Run ID: c5c8390e-d514-40b6-bd5d-b6c367afd440

Baseline: 153cd14
Comparison: 62f7f41
Diff

Optimization Goals: ✅ No significant changes detected

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	uds_dogstatsd_to_api_cpu	% cpu utilization	+0.65	[-0.20, +1.51]	1	Logs
➖	tcp_syslog_to_blackhole	ingress throughput	+0.65	[+0.59, +0.71]	1	Logs
➖	quality_gate_idle	memory utilization	+0.24	[+0.18, +0.29]	1	Logs bounds checks dashboard
➖	file_to_blackhole_1000ms_latency	egress throughput	+0.19	[-0.60, +0.97]	1	Logs
➖	file_to_blackhole_500ms_latency	egress throughput	+0.08	[-0.70, +0.86]	1	Logs
➖	quality_gate_idle_all_features	memory utilization	+0.05	[-0.01, +0.10]	1	Logs bounds checks dashboard
➖	file_to_blackhole_100ms_latency	egress throughput	+0.01	[-0.64, +0.66]	1	Logs
➖	file_to_blackhole_300ms_latency	egress throughput	+0.00	[-0.62, +0.63]	1	Logs
➖	file_to_blackhole_0ms_latency	egress throughput	+0.00	[-0.76, +0.77]	1	Logs
➖	tcp_dd_logs_filter_exclude	ingress throughput	+0.00	[-0.02, +0.02]	1	Logs
➖	file_to_blackhole_0ms_latency_http1	egress throughput	+0.00	[-0.81, +0.81]	1	Logs
➖	file_to_blackhole_0ms_latency_http2	egress throughput	-0.00	[-0.80, +0.80]	1	Logs
➖	uds_dogstatsd_to_api	ingress throughput	-0.00	[-0.27, +0.27]	1	Logs
➖	file_to_blackhole_1000ms_latency_linear_load	egress throughput	-0.27	[-0.74, +0.19]	1	Logs
➖	file_tree	memory utilization	-0.39	[-0.45, -0.33]	1	Logs
➖	quality_gate_logs	% cpu utilization	-0.40	[-3.28, +2.49]	1	Logs

Bounds Checks: ✅ Passed

perf	experiment	bounds_check_name	replicates_passed	links
✅	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	intake_connections	10/10	bounds checks dashboard
✅	quality_gate_idle	memory_usage	10/10	bounds checks dashboard
✅	quality_gate_idle_all_features	intake_connections	10/10	bounds checks dashboard
✅	quality_gate_idle_all_features	memory_usage	10/10	bounds checks dashboard
✅	quality_gate_logs	intake_connections	10/10
✅	quality_gate_logs	lost_bytes	10/10
✅	quality_gate_logs	memory_usage	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:

1. Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.

2. 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.

3. Its configuration does not mark it "erratic".

CI Pass/Fail Decision

✅ Passed. All Quality Gates passed.

• quality_gate_idle_all_features, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_idle_all_features, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_idle, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_idle, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check lost_bytes: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.

[grantseltzer]

I can't get the tests to actually run. Issues with running git:

=== RUN   TestIntegration
download /tmp/protobuf-integration-1060272402/src/google.golang.org/protobuf/.cache/protobuf-27.0
    integration_test.go:267: executing (git clone https://github.com/protocolbuffers/protobuf protobuf-27.0): exit status 128
        Cloning into 'protobuf-27.0'...
        error: RPC failed; curl 92 HTTP/2 stream 0 was not closed cleanly: CANCEL (err 8)
        error: 4217 bytes of body are still expected
        fetch-pack: unexpected disconnect while reading sideband packet
        fatal: early EOF
        fatal: fetch-pack: invalid index-pack output
--- FAIL: TestIntegration (118.60s)
FAIL
FAIL	google.golang.org/protobuf	120.028s
FAIL
    exploration_e2e_test.go:1038:
        	Error Trace:	/home/vagrant/datadog-agent/pkg/dynamicinstrumentation/testutil/exploration_e2e_test.go:1038
        	Error:      	Received unexpected error:
        	            	exit status 1
        	Test:       	TestExplorationGoDI
--- FAIL: TestExplorationGoDI (142.58s)
FAIL
exit status 1
FAIL	github.com/DataDog/datadog-agent/pkg/dynamicinstrumentation/testutil	142.615s

Regardless I think dependencies would be better served as git submodules, what do you think?.

[grantseltzer]

Can you document more the need to use signals for stopping/continuing procs?

My understanding so far is that we download the protobuf library, compile all test binaries, attach probes to all functions in the binaries, then run them and inspect results. Why is it necessary to start a process group and orchestrate with signals instead of running the binaries one at a time?

[GreenMatan]

In order to do what you suggest, I'll have to use Go DI in unusual way. With the Exploration Testing, I wanted to mimic real-world scenario as much as I can, including all the process tracking logic, analysis of binaries with their ProcessInfo struct, etc. Essentially treating Go DI and the project-under-test as two blackboxes.
Your recommendation could guide us in creating a new test that is tailored to unravel the scope of our support.

In the work of designing the moving parts of Exploration Testing I considered a couple of approaches and ended up going in the approach of running the tests in a blackbox without caring much about their nature. The infra is pretty flexible and can be easily adjusted to add more projects. I didn't want to manually import protobuf, look into how to run it continuously and be forced to put probes on predefined locations.
I do agree, though, that we have a challenge in determining which functions are actually executing. I don't think this challenge rule out the whole technique just yet. I have a few ideas of how to deal with that (use static analysis / profling to determine which symbols should be prioritized to probe)

Having said all of that, I still believe the exploration tests are incredibly useful in their current state. Simply running them helped me uncover 6+ crashes caused by nil/invalid indexing dereferences, 1 race condition, and 2 logical bugs - bugs that would have otherwise been really hard to find. Debugging it was also a bless. The fact that numerous subtests involve in the execution of protobuf, some of which are very short-lived, also helped uncover these edge cases.
I don't say that exploration tests are a complete task as they currently stand, but they're totally a tool to assess the readiness to private beta. We could choose to create more tests either by utilizing this infra or creating a new one, but the current exploration tests did prove themselves.

[grantseltzer]

1. Let's split this into two PRs, one for stability fixes and another for exploration testing.

2. Generally speaking i'd like to discuss making the testing set up simpler. Maybe we can start by having a service we own (or build on top of the existing sample service) which can use protobuf. We could even start multiple instances of it with a runner for the sake of stress testing. Each instance can even use multiple go routines for portions of the test. It would make it a lot easier to know what code paths are called (so we know if output percentages are accurate), and what compare event output we expect for accuracy.

[GreenMatan]

I've splitted the PR into Stability & Exploration Testing. The exploration testing PR is: #34423 and is stacked on top of this PR.

[agent-platform-auto-pr]

Test changes on VM

Use this command from test-infra-definitions to manually test this PR changes on a VM:

inv aws.create-vm --pipeline-id=56883287 --os-family=ubuntu

Note: This applies to commit 62f7f41

[agent-platform-auto-pr]

Static quality checks ✅

Please find below the results from static quality gates

Successful checks

Info

Result	Quality gate	On disk size	On disk size limit	On wire size	On wire size limit
✅	static_quality_gate_agent_deb_amd64	808.74MiB	841.59MiB	195.28MiB	210.72MiB
✅	static_quality_gate_agent_deb_arm64	799.62MiB	830.66MiB	176.82MiB	192.1MiB
✅	static_quality_gate_agent_rpm_amd64	808.88MiB	841.52MiB	198.18MiB	214.52MiB
✅	static_quality_gate_agent_rpm_arm64	799.67MiB	830.66MiB	178.77MiB	193.64MiB
✅	static_quality_gate_agent_suse_amd64	808.87MiB	841.52MiB	198.18MiB	214.52MiB
✅	static_quality_gate_agent_suse_arm64	799.67MiB	830.66MiB	178.77MiB	194.03MiB
✅	static_quality_gate_dogstatsd_deb_amd64	37.67MiB	49.7MiB	9.78MiB	20.6MiB
✅	static_quality_gate_dogstatsd_deb_arm64	36.27MiB	48.1MiB	8.48MiB	19.1MiB
✅	static_quality_gate_dogstatsd_rpm_amd64	37.67MiB	49.7MiB	9.79MiB	20.6MiB
✅	static_quality_gate_dogstatsd_suse_amd64	37.67MiB	49.7MiB	9.79MiB	20.6MiB
✅	static_quality_gate_iot_agent_deb_amd64	59.23MiB	69.0MiB	14.88MiB	24.8MiB
✅	static_quality_gate_iot_agent_deb_arm64	56.59MiB	66.4MiB	12.85MiB	22.8MiB
✅	static_quality_gate_iot_agent_rpm_amd64	59.23MiB	69.0MiB	14.9MiB	24.8MiB
✅	static_quality_gate_iot_agent_rpm_arm64	56.6MiB	66.4MiB	12.85MiB	22.8MiB
✅	static_quality_gate_iot_agent_suse_amd64	59.23MiB	69.0MiB	14.9MiB	24.8MiB
✅	static_quality_gate_docker_agent_amd64	893.2MiB	926.0MiB	298.81MiB	317.43MiB
✅	static_quality_gate_docker_agent_arm64	907.41MiB	939.07MiB	284.54MiB	302.43MiB
✅	static_quality_gate_docker_agent_jmx_amd64	1.07GiB	1.1GiB	373.9MiB	392.54MiB
✅	static_quality_gate_docker_agent_jmx_arm64	1.07GiB	1.1GiB	355.6MiB	373.21MiB
✅	static_quality_gate_docker_dogstatsd_amd64	45.81MiB	57.88MiB	17.28MiB	28.29MiB
✅	static_quality_gate_docker_dogstatsd_arm64	44.45MiB	56.27MiB	16.16MiB	27.06MiB
✅	static_quality_gate_docker_cluster_agent_amd64	264.93MiB	274.78MiB	106.35MiB	116.28MiB
✅	static_quality_gate_docker_cluster_agent_arm64	280.89MiB	290.82MiB	101.18MiB	111.12MiB

[grantseltzer]

There's other fields which are set in procInfo from AnalyzeBinary besides the type map, so i'm not sure if this is the correct way to do this check.

I originally thought that it would be an issue because when the binary is inspected we're only analyzing types for target functions, so if a config is updated, we wouldn't be able to parse types for the new configuration. However testing this branch doesn't show any issues.

What were the reservations you had about this particular change?

[grantseltzer]

Ok I did further investigation and realized this code doesn't belong here at all. AnalyzeBinary needs to be called for the sake of the RC config manager, but the memory config manager (i.e. what file based configuration is built on) has separate callbacks for when procs/configs are updated which call AnalyzeBinary already.

So remove this call to AnalyzeBinary all together and add it before the call to applyConfigUpdate() on line 239 (without the condition nil check)