out_stackdriver: fix multi-worker resource-metadata data race (SIGSEGV)

[erain]

Summary

Fixes a use-after-free / double-free crash (SIGSEGV) in the Stackdriver output when it runs with workers >= 2.

stackdriver_format() extracts per-record Kubernetes resource metadata (pod_name, namespace_name, container_name, node_name, local_resource_id) and stored it on the shared plugin context (struct flb_stackdriver), repeatedly flb_sds_destroy() + flb_sds_create()-ing those fields. With multiple flush worker threads, two concurrent flushes race on the same context strings — freeing and reading the same buffer at once — a data race that manifests as heap corruption / SIGSEGV under load.

This moves those fields into a per-flush stack-local struct stackdriver_format_ctx owned by stackdriver_format(), threads the resource helpers through it, and frees them in a single cleanup path. The shared context no longer holds mutable per-record state, so the format path is correct for any worker count without locking.

Testing

• New runtime test resource_k8s_container_concurrency (2 workers × 5 concurrent input streams); ctest -R out_stackdriver passes.
• Verified under ThreadSanitizer: the resource-metadata data races reported with workers=2 before this change are eliminated.
• [N/A] Valgrind: covered by the runtime test suite.

Documentation

• [N/A] No user-facing or behavioral change.

Backporting

• Candidate for backport (crash fix).

---

Fluent Bit is licensed under Apache 2.0, by submitting this pull request I understand that this code will be released under the terms of that license.

Summary by CodeRabbit

• Bug Fixes
  • Improved Stackdriver Kubernetes log formatting by extracting and applying the correct local resource ID per entry, making namespace/pod/container/node resource labels more consistent and preventing cross-entry mix-ups.
  • Made payload key matching case-insensitive for more tolerant field handling.
  • Refined formatter assembly and cleanup so empty/failed batches are handled more reliably under load.
• Tests
  • Added a high-volume concurrency test for k8s_container monitored resource formatting.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 77a292bc-8e09-4bc0-83f0-192ec9a47c26

📥 Commits

Reviewing files that changed from the base of the PR and between eb0694a and cd418bc.

📒 Files selected for processing (1)

• plugins/out_stackdriver/stackdriver.c

🚧 Files skipped from review as they are similar to previous changes (1)

• plugins/out_stackdriver/stackdriver.c

---

📝 Walkthrough

Walkthrough

The Stackdriver output now extracts Kubernetes labels into a per-call format context, removes those fields from shared plugin state, routes formatting failures through consolidated cleanup, switches msgpack key matching to case-insensitive comparison, and adds a runtime concurrency test.

Changes

Stackdriver per-call format context refactor

Layer / File(s)	Summary
Format context struct and shared state removal plugins/out_stackdriver/stackdriver.c, plugins/out_stackdriver/stackdriver.h, plugins/out_stackdriver/stackdriver_conf.c	A per-call format context struct is added, and Kubernetes label fields are removed from the shared Stackdriver struct and its destroy path.
Resource ID extraction and label parsing plugins/out_stackdriver/stackdriver.c	Resource ID extraction, monitored-resource label selection, regex matching, and regex callbacks are updated to read and write Kubernetes label values through the per-call format context.
stackdriver_format() cleanup and payload packing plugins/out_stackdriver/stackdriver.c	stackdriver_format() now initializes and cleans up a local format context, routes Kubernetes failures through a shared cleanup path, builds output entries incrementally, and packs Kubernetes resource labels from that context; get_msgpack_obj() uses case-insensitive key matching and pack_payload() reorganizes special-field branches.
Concurrency runtime test tests/runtime/out_stackdriver.c	A runtime test adds five concurrent lib inputs, pushes repeated JSON payloads with logging.googleapis.com/local_resource_id, and registers the new case in TEST_LIST.

Estimated code review effort: 4 (Complex) | ~60 minutes

Possibly related PRs

• fluent/fluent-bit#11879: Both PRs modify stackdriver_format()'s error cleanup and label-handling flow in the same function.

Suggested reviewers: braydonk, cosmo0920, JeffLuoo

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: fixing a multi-worker Stackdriver resource-metadata race causing SIGSEGV.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 2

🧹 Nitpick comments (1)

plugins/out_stackdriver/stackdriver.c (1)

729-730: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Align the changed C style with this repository.

Line 730 puts a function opening brace on the signature line, and Line 1080 uses a // comment.

Proposed style cleanup

 static int extract_local_resource_id(const void *data, size_t bytes,
-                                     struct stackdriver_format_ctx *fmt_ctx, const char *tag) {
+                                     struct stackdriver_format_ctx *fmt_ctx,
+                                     const char *tag)
+{
@@
-    // parsing local_resource_id from tag takes higher priority
+    /* Parsing local_resource_id from tag takes higher priority. */

As per coding guidelines, C files should put function opening braces on the next line and use /* ... */ comments.

Also applies to: 1080-1080

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@plugins/out_stackdriver/stackdriver.c` around lines 729 - 730, Update the C
style in stackdriver.c to match the repository conventions: move the opening
brace of extract_local_resource_id onto the next line instead of keeping it on
the function signature line, and replace the single-line // comment near the
referenced logging/comment block with a /* ... */ style comment. Keep the same
behavior while adjusting only the formatting in the affected function and
comment site.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@plugins/out_stackdriver/stackdriver.c`:
- Around line 1020-1022: Guard the `local_resource_id` slicing in the
Stackdriver matching logic before using `ctx->tag_prefix` as an offset. In the
block that assigns `prefix_len`, `str_to_be_matcheds`, and `len_to_be_matched`,
add a length check on `fmt_ctx->local_resource_id` against `prefix_len` so
shorter values are rejected or skipped before pointer arithmetic and regex
matching. Use the existing `ctx->tag_prefix` and `fmt_ctx->local_resource_id`
handling in this section to keep the fix localized and prevent advancing past
the SDS buffer.

In `@tests/runtime/out_stackdriver.c`:
- Around line 3687-3698: The stackdriver runtime test setup is bypassing the
formatter hook, so it reaches the real flush/auth path instead of exercising
stackdriver_format() in-process. Update the test block around
flb_output()/flb_output_set()/flb_start() to call flb_output_set_test() before
starting the engine, using the same stackdriver output instance so the formatter
path is covered and the unused label assertions remain meaningful.

---

Nitpick comments:
In `@plugins/out_stackdriver/stackdriver.c`:
- Around line 729-730: Update the C style in stackdriver.c to match the
repository conventions: move the opening brace of extract_local_resource_id onto
the next line instead of keeping it on the function signature line, and replace
the single-line // comment near the referenced logging/comment block with a /*
... */ style comment. Keep the same behavior while adjusting only the formatting
in the affected function and comment site.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: b095ec6f-5b28-48f5-9cd9-5eadf633baf7

📥 Commits

Reviewing files that changed from the base of the PR and between b81c252 and eb0694a.

📒 Files selected for processing (4)

• plugins/out_stackdriver/stackdriver.c
• plugins/out_stackdriver/stackdriver.h
• plugins/out_stackdriver/stackdriver_conf.c
• tests/runtime/out_stackdriver.c

💤 Files with no reviewable changes (1)

• plugins/out_stackdriver/stackdriver.h