Repeated crashes due to SIGSEGV in mimalloc during concurrent federated search formatting and indexing (v1.51, Linux AArch64)

Describe the bug We’ve had Meilisearch crash at least three times in production after upgrading from a v1.3x release to v1.51.0.

Each time, systemd reports a native segfault:

Main process exited, code=dumped, status=11/SEGV

The core dumps point to the same general path: mimalloc while Meilisearch is formatting results for a federated search.

mi_heap_malloc_zero_aligned_at_generic
MatchesIter::next
Matcher::compute_matches
Matcher::format
format_fields
HitMaker::make_hit
SearchByIndex::execute

Other threads were indexing documents and writing to LMDB at the same time. This doesn’t appear to be an OOM issue. The server has 64 GiB of RAM, little swap usage, and no corresponding kernel OOM event.

To Reproduce We don’t have a reliable standalone reproduction yet. It happens under normal production traffic with indexing and federated searches running concurrently. Indexing is limited to two threads and 12 GiB of memory.

Expected behavior Concurrent indexing and searching should not terminate the Meilisearch process.

Screenshots If applicable, add screenshots to help explain your problem.

Meilisearch version: v1.51.0, upgraded from a v1.3x release. The crashes continue after the upgrade, EC2 resize from 32 to 64 GiB RAM, and EBS expansion from 120 to 300 GiB.

Additional context

• AWS EC2, AArch64
• Ubuntu 24.04
• 8 vCPUs and 64 GiB RAM
• EBS was increased from 120 to 300 GiB
• Two crashes have very similar mimalloc/search-formatting traces
• The process restarts normally afterward
• We have the full symbolized backtraces and preserved cores. We can share the cores privately since they may contain application data.