[CORE-15249] sr/json: Optimise is_superset to increase recursion depth by BenPope · Pull Request #29290 · redpanda-data/redpanda

BenPope · 2026-01-16T13:06:35Z

Optimise compatibility checking for json to allow more nested schemas.

The depth was tuned by adding these options to the compiler to determine the stack frame sizes:

--copt=-fstack-usage --copt=-Wframe-larger-than=512 --copt=-Wno-error=frame-larger-than

Backports Required

Release Notes

Improvements

Schema Registry: Optimise compatibility checking for json to allow more nested schemas.

Copilot

Pull request overview

This PR optimizes JSON schema compatibility checking to support deeper schema nesting by reducing recursion depth and improving the efficiency of type checking operations. The primary changes replace recursive function calls with iterative approaches and optimize data structures used in the type comparison logic.

Changes:

Introduced iterative schema traversal to prevent stack overflow on deeply nested schemas
Replaced type list with bitset-based operations for more efficient type comparisons
Refactored helper functions to reduce stack frame overhead during formatting operations

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/v/pandaproxy/schema_registry/json.cc	Converted recursive schema collection to iterative approach, replaced type list operations with bitset operations, optimized formatting helpers
src/v/pandaproxy/schema_registry/test/test_json_schema.cc	Added test for deeply nested schemas to validate stack overflow prevention

Copilot · 2026-01-16T13:07:08Z

-                    throw as_exception(invalid_schema(
-                      fmt::format("{} has more than one combinator", pj{v})));
+                    throw_invalid_schema(
+                      "schema_invalid",


The first argument "schema_invalid" should not be passed to throw_invalid_schema. Based on the function signature at lines 73-75, throw_invalid_schema takes a format string and arguments, but the format string here is "{} has more than one combinator" (second argument). The extra "schema_invalid" string will cause a format string mismatch. This differs from the pattern used elsewhere where throw_invalid_schema is called with only the format string and arguments.

Suggested change

"schema_invalid",

IoannisRP

🔥🔥🔥🔥🔥

BenPope and claude committed...

Excellent way to shift blame if things break 😆

Really nice PR! a couple of minor comments

IoannisRP · 2026-01-16T13:22:06Z

+    // Setting the limit to 66 causes corruption of the heap stack due to stack
    // overflow.
-    constexpr int max_test_depth = 31;
+    constexpr int max_test_depth = 65;


chore:

you say 100 to commit message. Also in the messages of the previous commit it mentions much bigger depths.

For future clarity, i say just move this PR first in the commit history and increase depth to something unreasonable (like 1000), to make clear that this is just for testing purposes and it will be undone.

Instead of changing it in every PR, you can just document new limit in the comment... and then at the end, set it to whatever is needed. The limit is kinda dependent on platform, anyways.

chore:

you say 100 to commit message. Also in the messages of the previous commit it mentions much bigger depths.

I shouldn't have reordered commits! Fix incoming

For future clarity, i say just move this PR first in the commit history and increase depth to something unreasonable (like 1000), to make clear that this is just for testing purposes and it will be undone.

Instead of changing it in every PR, you can just document new limit in the comment... and then at the end, set it to whatever is needed. The limit is kinda dependent on platform, anyways.

The problem is that we never really know what the max is, the heap corruption may appear at shutdown several depths higher.

IoannisRP · 2026-01-16T13:50:06Z

    // Setting the limit to 84 causes corruption of the heap stack due to stack
    // overflow.
-    constexpr int max_test_depth = 83;
+    constexpr int max_test_depth = 96;


nit:

comment was not updated

IoannisRP · 2026-01-16T13:54:26Z

 namespace {

+// Helper to create schema errors without inlining fmt::format machinery
+// into the caller's stack frame. Marked cold since errors are exceptional.


Marked cold since errors are exceptional

Is there something in the code that actually marks them as cold? 🤔
I was expecting to see a noinline or something like that

Marked cold since errors are exceptional

Is there something in the code that actually marks them as cold? 🤔 I was expecting to see a noinline or something like that

I removed them, it made no difference.

IoannisRP · 2026-01-16T14:00:50Z

I think it would also be useful to add before/after stack frame sizes on the cover (once you are ready to commit, as they might still change will working through it)

vbotbuildovich · 2026-01-16T14:31:13Z

CI test results

test results on build#79153

test_class	test_method	test_arguments	test_kind	job_url	test_status	passed	reason	test_history
MountUnmountIcebergTest	test_simple_remount	{"cloud_storage_type": 1}	integration	https://buildkite.com/redpanda/redpanda/builds/79153#019bc6fb-2351-4aff-b1d5-aa1528ca5944	FLAKY	8/11	Test PASSES after retries.No significant increase in flaky rate(baseline=0.1833, p0=0.5718, reject_threshold=0.0100. adj_baseline=0.4553, p1=0.0938, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=MountUnmountIcebergTest&test_method=test_simple_remount
src/v/pandaproxy/schema_registry/test/test_json_schema	src/v/pandaproxy/schema_registry/test/test_json_schema		unit	https://buildkite.com/redpanda/redpanda/builds/79153#019bc6eb-024a-4f12-a374-c2f601f5c015	FAIL	0/1

test results on build#79165

test_class	test_method	test_arguments	test_kind	job_url	test_status	passed	reason	test_history
MountUnmountIcebergTest	test_simple_remount	{"cloud_storage_type": 1}	integration	https://buildkite.com/redpanda/redpanda/builds/79165#019bc76d-b8a4-404c-b19f-c91e1742916b	FLAKY	9/11	Test PASSES after retries.No significant increase in flaky rate(baseline=0.1830, p0=0.8675, reject_threshold=0.0100. adj_baseline=0.4547, p1=0.0217, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=MountUnmountIcebergTest&test_method=test_simple_remount
NodesDecommissioningTest	test_decommission_status	null	integration	https://buildkite.com/redpanda/redpanda/builds/79165#019bc767-d5aa-455a-ba9d-5df569322a36	FLAKY	10/11	Test PASSES after retries.No significant increase in flaky rate(baseline=0.0372, p0=1.0000, reject_threshold=0.0100. adj_baseline=0.1076, p1=0.3202, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=NodesDecommissioningTest&test_method=test_decommission_status

test results on build#79186

test_class	test_method	test_arguments	test_kind	job_url	test_status	passed	reason	test_history
WriteCachingFailureInjectionE2ETest	test_crash_all	{"use_transactions": false}	integration	https://buildkite.com/redpanda/redpanda/builds/79186#019bc7e9-1a03-46e7-805e-bf63485ae56b	FLAKY	8/11	Test PASSES after retries.No significant increase in flaky rate(baseline=0.1060, p0=0.2873, reject_threshold=0.0100. adj_baseline=0.2856, p1=0.4222, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=WriteCachingFailureInjectionE2ETest&test_method=test_crash_all
TxAtomicProduceConsumeTest	test_basic_tx_consumer_transform_produce	{"with_failures": true}	integration	https://buildkite.com/redpanda/redpanda/builds/79186#019bc7e9-19ff-4176-9834-2c4bd5923a51	FLAKY	10/11	Test PASSES after retries.No significant increase in flaky rate(baseline=0.0041, p0=1.0000, reject_threshold=0.0100. adj_baseline=0.1000, p1=0.3487, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=TxAtomicProduceConsumeTest&test_method=test_basic_tx_consumer_transform_produce

test results on build#79249

test_class	test_method	test_arguments	test_kind	job_url	test_status	passed	reason	test_history
ControllerForcedReconfiguration_Size5	test_cluster_recovery	{"scenario": "Simple"}	integration	https://buildkite.com/redpanda/redpanda/builds/79249#019bd639-9dcf-4560-b184-b19fc795e8f4	FLAKY	10/11	Test PASSES after retries.No significant increase in flaky rate(baseline=0.0315, p0=1.0000, reject_threshold=0.0100. adj_baseline=0.1000, p1=0.3487, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=ControllerForcedReconfiguration_Size5&test_method=test_cluster_recovery
DatalakeCustomPartitioningTest	test_sticky_default	{"catalog_type": "rest_jdbc", "cloud_storage_type": 1}	integration	https://buildkite.com/redpanda/redpanda/builds/79249#019bd639-9dcf-4560-b184-b19fc795e8f4	FLAKY	10/11	Test PASSES after retries.No significant increase in flaky rate(baseline=0.0000, p0=1.0000, reject_threshold=0.0100. adj_baseline=0.1000, p1=0.3487, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=DatalakeCustomPartitioningTest&test_method=test_sticky_default
NodesDecommissioningTest	test_decommission_status	null	integration	https://buildkite.com/redpanda/redpanda/builds/79249#019bd63e-9ad2-4301-9e51-7d0d51f0dc02	FLAKY	19/21	Test PASSES after retries.No significant increase in flaky rate(baseline=0.0387, p0=0.5462, reject_threshold=0.0100. adj_baseline=0.1118, p1=0.3287, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=NodesDecommissioningTest&test_method=test_decommission_status
RedpandaNodeOperationsSmokeTest	test_node_ops_smoke_test	{"cloud_storage_type": 1, "mixed_versions": false}	integration	https://buildkite.com/redpanda/redpanda/builds/79249#019bd63e-9ada-4e8c-800f-a930b9c5d2c5	FAIL	0/1		https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=RedpandaNodeOperationsSmokeTest&test_method=test_node_ops_smoke_test
SimpleEndToEndTest	test_relaxed_acks	{"write_caching": false}	integration	https://buildkite.com/redpanda/redpanda/builds/79249#019bd63e-9adc-4d39-8934-ad18f786f7ff	FLAKY	10/11	Test PASSES after retries.No significant increase in flaky rate(baseline=0.0134, p0=1.0000, reject_threshold=0.0100. adj_baseline=0.1000, p1=0.3487, trust_threshold=0.5000)	https://redpanda.metabaseapp.com/dashboard/87-tests?tab=142-dt-individual-test-history&test_class=SimpleEndToEndTest&test_method=test_relaxed_acks

michael-redpanda

dang this looks cool

michael-redpanda · 2026-01-16T14:29:16Z

-  jsoncons::jsonpointer::json_pointer this_obj_ptr,
-  jsoncons::ojson& this_obj,
-  json_schema_dialect dialect) {
+  std::vector<collect_work_item>& work_stack,


Wouldn't we run into a oversized alloc issue here using a std::vector?

Of note: I've found that Claude really needs to be reminded about which containers to use

michael-redpanda · 2026-01-16T14:30:09Z

+  jsoncons::jsonpointer::json_pointer obj_ptr,
+  jsoncons::ojson& obj,
+  json_schema_dialect dialect) {
+    std::vector<collect_work_item> work_stack;


same issue here

BenPope · 2026-01-16T15:13:51Z

Changes in force-push

Address review comments
Don't move from a const&
Add cycle detection
Increase max_recursion_depth

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

Copilot · 2026-01-16T15:14:58Z

+    static constexpr int max_recursion_depth{200};
    int _ref_units{max_recursion_depth};


The max_recursion_depth has been increased from 8 to 200, but this constant is now misleading since the main recursion prevention mechanism is the cycle detection via visited set. Consider renaming this constant to better reflect its purpose (e.g., max_ref_resolution_depth) or documenting why this specific limit remains necessary alongside cycle detection.

Suggested change

static constexpr int max_recursion_depth{200};

int _ref_units{max_recursion_depth};

// Safety cap on the depth/amount of $ref resolution. Cycle detection via

// the visited set is the primary recursion guard; this remains as a

// secondary limit to prevent pathological inputs from consuming

// unbounded resources.

static constexpr int max_ref_resolution_depth{200};

int _ref_units{max_ref_resolution_depth};

Copilot · 2026-01-16T15:14:58Z

 };
-// enough inlined space to hold all the values of json_type
-using json_type_list = absl::InlinedVector<json_type, 7>;
+using json_type_list = enum_bitset<json_type, 7>;


The size parameter 7 is a magic number that should be replaced with a constant or derived from the enum. Consider using static_cast<std::size_t>(json_type::null) + 1 or defining a json_type_count constant to make the relationship between the enum and bitset size explicit.

Copilot · 2026-01-16T15:14:58Z

+  jsoncons::uri base_uri,
+  jsoncons::jsonpointer::json_pointer obj_ptr,
+  jsoncons::ojson& obj,
+  json_schema_dialect dialect) {


The variable is named work_stack but uses a chunked_vector. Consider renaming to work_queue or work_list since vectors don't necessarily convey LIFO semantics, or add a comment explaining why a vector is used for stack operations.

Suggested change

json_schema_dialect dialect) {

json_schema_dialect dialect) {

// Use chunked_vector as an explicit LIFO stack (push_back/back/pop_back)

// to avoid recursion on deeply nested schemas.

pgellert

Great stuff! Can you please add the commands that you used for deriving the stack sizes / warning to the PR cover letter for future reference?

pgellert · 2026-01-16T15:07:01Z

+    // Not implemented - only needed for input streams
+    char Peek() const { return 0; }
+    char Take() { return 0; }
+    size_t Tell() const { return 0; }
+    char* PutBegin() { return nullptr; }
+    size_t PutEnd(char*) { return 0; }


Should these throw unimplemented / assert unreachable?

Bump on this. Mainly to:

Confirm these are not accidentally used now

Ensure they are not accidentally used in the future

Show that they are unused if we need to trace down bugs in the future in this area

Add a parameterized test that validates stack frame size optimizations by testing compatibility checking on deeply nested JSON schemas. The test creates schemas with nested object properties and measures the maximum depth achievable before stack overflow. Current baseline with optimizations: depth 70 (with jsoncons validation disabled and max_recursion_depth=100). Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Temporarily disable jsoncons metaschema validation and increase max_recursion_depth to enable stack depth testing. This allows the test_object_recursion_depths test to measure the actual stack limits of the is_superset code path without being limited by jsoncons. TODO: Re-enable validation once stack optimizations are complete. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Stack frame size changes: - ::(anonymous namespace)::is_superset: 1464 -> 1368 (-96 bytes) Signed-off-by: Ben Pope <ben@redpanda.com>

Two improvements from passing std::filesystem::path by const reference: is_superset: 1368 → 1080 (-288 bytes) is_object_properties_superset: 712 → 648 (-64 bytes) Signed-off-by: Ben Pope <ben@redpanda.com>

This allows processing deeply nested schemas without stack overflow. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…_value_superset Pass json_incompatibility_type and path separately instead of pre-constructed json_incompatibility objects. The path is only appended with prop_name when an error actually occurs, avoiding temporary filesystem::path construction in the caller's stack frame. Stack frame size changes: - is_numeric_superset: 1032 -> 680 (-352 bytes) - is_array_superset: 856 -> 616 (-240 bytes) - is_object_superset: 1032 -> 792 (-240 bytes) - is_string_superset: 664 -> <512 (no longer triggers warning) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

BenPope · 2026-01-16T17:32:36Z

Changes in force-push

Don't pass compatibility_context by reference into is_superset.
Tweak some limits

IoannisRP

My only concern is that I don't see any tests for

sr/json: Add cycle detection

Do we have any existing tests that cover this?

IoannisRP · 2026-01-19T11:26:45Z

+template<typename... Args>
+error_info
+make_invalid_schema(fmt::format_string<Args...> fmt, Args&&... args) {
+    return error_info{
+      error_code::schema_invalid,
+      fmt::format(fmt, std::forward<Args>(args)...)};
+}
+
+template<typename... Args>
+[[noreturn]] void
+throw_invalid_schema(fmt::format_string<Args...> fmt, Args&&... args) {
+    throw as_exception(make_invalid_schema(fmt, std::forward<Args>(args)...));
+}


nit:
maybe add a noinline attribute on these for future-proofing the not-inlined behavior

Suggested change

template<typename... Args>

error_info

make_invalid_schema(fmt::format_string<Args...> fmt, Args&&... args) {

return error_info{

error_code::schema_invalid,

fmt::format(fmt, std::forward<Args>(args)...)};

}

template<typename... Args>

[[noreturn]] void

throw_invalid_schema(fmt::format_string<Args...> fmt, Args&&... args) {

throw as_exception(make_invalid_schema(fmt, std::forward<Args>(args)...));

}

template<typename... Args>

[[clang::noinline]] error_info

make_invalid_schema(fmt::format_string<Args...> fmt, Args&&... args) {

return error_info{

error_code::schema_invalid,

fmt::format(fmt, std::forward<Args>(args)...)};

}

template<typename... Args>

[[noreturn]][[clang::noinline]] void

throw_invalid_schema(fmt::format_string<Args...> fmt, Args&&... args) {

throw as_exception(make_invalid_schema(fmt, std::forward<Args>(args)...));

}

It's not obvious that the noinline helps - I'm sure I tested with and without, and there was no difference.

the only reason for these functions to help is for them to not be inlined, right?

If so, the tag indeed would not "help". But what is inlined might change as we change compiler versions. Adding the tag makes it explicit and documents the desired behavior.

IoannisRP · 2026-01-19T11:29:24Z

-    // overflow, which manifests as a crash during Seastar shutdown.
-    constexpr int max_test_depth = 129;
+    // Note: jsoncons validation overflows the stack at about 31.
+    // With validation disabled, setting the limit above ~139 causes corruption


130 became 139 in comment. Was this on purpose?

130 became 139 in comment. Was this on purpose?

No

BenPope · 2026-01-19T11:36:35Z

My only concern is that I don't see any tests for

sr/json: Add cycle detection

Do we have any existing tests that cover this?

The commit changes a test for cycles to now succeed rather than fail, and adds a comment along those lines.

pgellert · 2026-01-19T11:17:16Z

+// Helper to create schema errors without inlining fmt::format machinery
+// into the caller's stack frame.
+template<typename... Args>
+error_info
+make_invalid_schema(fmt::format_string<Args...> fmt, Args&&... args) {


Does this need a [[noinline]]? Otherwise, I'd guess whether this + fmt::format gets inlined only depends on how aggressively the compiler is trying to optimize.

I tried with and without and there was no change to the frame size

pgellert · 2026-01-19T11:35:10Z

+    // Not implemented - only needed for input streams
+    char Peek() const { return 0; }
+    char Take() { return 0; }
+    size_t Tell() const { return 0; }
+    char* PutBegin() { return nullptr; }
+    size_t PutEnd(char*) { return 0; }


Bump on this. Mainly to:

Confirm these are not accidentally used now

Ensure they are not accidentally used in the future

Show that they are unused if we need to trace down bugs in the future in this area

Add throw_as_exception helper to move fmt::format machinery out of callers' stack frames. Replace pj/pjp ostream formatters with format_to() methods that write directly to fmt::iterator. Stack frame size changes: - get_schema: 712 -> 600 (-112 bytes) - is_numeric_superset: 712 -> 680 (-32 bytes) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

This is the limit before a test fails. Signed-off-by: Ben Pope <ben@redpanda.com>

Signed-off-by: Ben Pope <ben@redpanda.com>

Track visited (older, newer) schema pairs for cycle detection. If we encounter the same pair twice during traversal, we've found a cycle and can return early (compatible by assumption - cycles that differ would have been caught on first visit). Signed-off-by: Ben Pope <ben@redpanda.com>

vbotbuildovich · 2026-01-19T13:23:00Z

Retry command for Build#79249

please wait until all jobs are finished before running the slash command

/ci-repeat 1
skip-redpanda-build
skip-units
skip-rebase
tests/rptest/tests/random_node_operations_smoke_test.py::RedpandaNodeOperationsSmokeTest.test_node_ops_smoke_test@{"cloud_storage_type":1,"mixed_versions":false}

BenPope · 2026-01-19T14:01:05Z

/ci-repeat 1
skip-redpanda-build
skip-units
skip-rebase
tests/rptest/tests/random_node_operations_smoke_test.py::RedpandaNodeOperationsSmokeTest.test_node_ops_smoke_test@{"cloud_storage_type":1,"mixed_versions":false}

vbotbuildovich · 2026-01-19T14:51:39Z

/backport v25.3.x

vbotbuildovich · 2026-01-19T14:52:59Z

Failed to create a backport PR to v25.3.x branch. I tried:

git remote add upstream https://github.com/redpanda-data/redpanda.git
git fetch --all
git checkout -b backport-pr-29290-v25.3.x-892 remotes/upstream/v25.3.x
git cherry-pick -x 01b3cdc3d8 4012cdf270 8d40d84264 4c89a3c65b 3fb5a8010e 5d96e842b1 219f6c8c96 542ef24a70 543b1f422e fb6e585765 86c01d610d

Workflow run logs.

BenPope requested review from IoannisRP and pgellert January 16, 2026 13:06

BenPope self-assigned this Jan 16, 2026

Copilot AI review requested due to automatic review settings January 16, 2026 13:06

github-actions Bot added the area/redpanda label Jan 16, 2026

Copilot AI reviewed Jan 16, 2026

View reviewed changes

IoannisRP reviewed Jan 16, 2026

View reviewed changes

michael-redpanda reviewed Jan 16, 2026

View reviewed changes

michael-redpanda requested a review from nguyen-andrew January 16, 2026 14:33

BenPope force-pushed the CORE-15249/sr/json_recursion_depth_2 branch from 0b93042 to 8eb84fd Compare January 16, 2026 15:11

github-actions Bot added the area/build label Jan 16, 2026

BenPope requested review from IoannisRP, Copilot and michael-redpanda January 16, 2026 15:13

Copilot AI reviewed Jan 16, 2026

View reviewed changes

pgellert reviewed Jan 16, 2026

View reviewed changes

BenPope and others added 6 commits January 16, 2026 16:14

sr/json: Replace InlinedVector with std::bitset

8d40d84

Stack frame size changes: - ::(anonymous namespace)::is_superset: 1464 -> 1368 (-96 bytes) Signed-off-by: Ben Pope <ben@redpanda.com>

sr/json: Reduce stack by passing path by ref

4c89a3c

Two improvements from passing std::filesystem::path by const reference: is_superset: 1368 → 1080 (-288 bytes) is_object_properties_superset: 712 → 648 (-64 bytes) Signed-off-by: Ben Pope <ben@redpanda.com>

sr/json: Make collect_bundled_schemas_and_fix_refs iterative

3fb5a80

This allows processing deeply nested schemas without stack overflow. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

BenPope force-pushed the CORE-15249/sr/json_recursion_depth_2 branch from 8eb84fd to 82ed17a Compare January 16, 2026 17:28

BenPope requested a review from pgellert January 16, 2026 17:32

IoannisRP reviewed Jan 19, 2026

View reviewed changes

pgellert previously approved these changes Jan 19, 2026

View reviewed changes

BenPope and others added 5 commits January 19, 2026 12:15

sr/json: Increase max_recursion_depth

542ef24

This is the limit before a test fails. Signed-off-by: Ben Pope <ben@redpanda.com>

sr/json: reenable validation

543b1f4

Signed-off-by: Ben Pope <ben@redpanda.com>

sr/json: Prefer chunked_hash_map

fb6e585

Signed-off-by: Ben Pope <ben@redpanda.com>

BenPope dismissed pgellert’s stale review via 86c01d6 January 19, 2026 12:16

BenPope force-pushed the CORE-15249/sr/json_recursion_depth_2 branch from 82ed17a to 86c01d6 Compare January 19, 2026 12:16

BenPope requested review from IoannisRP and pgellert January 19, 2026 12:16

pgellert approved these changes Jan 19, 2026

View reviewed changes

IoannisRP approved these changes Jan 19, 2026

View reviewed changes

BenPope merged commit 9065eef into redpanda-data:dev Jan 19, 2026
19 checks passed

vbotbuildovich mentioned this pull request Jan 19, 2026

[v25.3.x] [CORE-15249] sr/json: Optimise is_superset to increase recursion depth #29315

Closed

BenPope mentioned this pull request Jan 19, 2026

[v25.3.x] [CORE-15249] sr/json: Optimise is_superset to increase recursion depth #29317

Merged

pgellert mentioned this pull request Feb 6, 2026

schema_registry/test: lower json recursion depth locally #29546

Merged

7 tasks

		static constexpr int max_recursion_depth{200};
		int _ref_units{max_recursion_depth};

-    static constexpr int max_recursion_depth{200};
-    int _ref_units{max_recursion_depth};
+    // Safety cap on the depth/amount of $ref resolution. Cycle detection via
+    // the visited set is the primary recursion guard; this remains as a
+    // secondary limit to prevent pathological inputs from consuming
+    // unbounded resources.
+    static constexpr int max_ref_resolution_depth{200};
+    int _ref_units{max_ref_resolution_depth};

Conversation

BenPope commented Jan 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Backports Required

Release Notes

Improvements

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

IoannisRP left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

IoannisRP commented Jan 16, 2026

Uh oh!

vbotbuildovich commented Jan 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CI test results

Uh oh!

michael-redpanda left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

BenPope commented Jan 16, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

pgellert left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pgellert Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

BenPope commented Jan 16, 2026

Uh oh!

IoannisRP left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

BenPope commented Jan 16, 2026 •

edited

Loading

vbotbuildovich commented Jan 16, 2026 •

edited

Loading

pgellert Jan 19, 2026 •

edited

Loading

pgellert Jan 19, 2026 •

edited

Loading