Window Fuzzer crashes with SIGSEGV (null pointer dereference)

[mbasmanova]

The Window Fuzzer crashes with a segfault (SIGSEGV, null pointer dereference at address 0x0) early in the fuzzer run.

Example failure: https://github.com/facebookincubator/velox/actions/runs/23534784917/job/68510203575

Instance 1 (seed 454204341) crashed within ~5 seconds of starting. Instances 2, 3, and 4 passed.

Details

Timeline from the logs:

1. Presto is up and running (CREATE SCHEMA succeeds)
2. All 4 fuzzer instances start simultaneously
3. Instance 1 hits a CURL timeout on fetchNext (PrestoQueryRunner.cpp:580) — query was submitted but fetching results timed out, likely due to resource contention from 4 concurrent instances
4. The timeout exception is caught and handled (returned as kReferenceQueryFail)
5. Next iteration begins — zstd compressor is initialized (during writeToFile)
6. SIGSEGV — null pointer dereference (address not mapped to object)

E20260325 03:13:26.273557   501 Exceptions.h:87] Line: .../PrestoQueryRunner.cpp:580, Function:fetchNext,
  Expression: CURLE_OK == res (0 vs. 28) Get request failed: Timeout was reached
I0325 03:13:26.286706   501 Compression.cpp:660] Initialized zstd compressor with compression level 7
*** Signal 11 (SIGSEGV) (0x0) received by PID 501 ... (code: address not mapped to object), stack trace: ***
(error retrieving stack trace)

Stack trace was not retrievable. The crash is a real bug — regardless of what caused the CURL timeout, the code should not segfault.

Repro

# Start a local Presto server, then run:
./velox_window_fuzzer_test \
    --seed 454204341 \
    --duration_sec 10 \
    --batch_size 50 \
    --minloglevel 0 \
    --enable_window_reference_verification \
    --presto_url=http://127.0.0.1:8080

[mbasmanova]

CC: @aditi-pandit @kagamiori @peterenescu @kgpai @pratikpugalia