SIGSEGV 0x559e755018d0 (unknown) dfly::detail::ascii_unpack_simd()

[jdheyburn]

Describe the bug

We encountered a few times a SIGSEGV crash a few times, this was against a large dataset of 31GiB of fake data during load testing.

First occurrence

*** SIGSEGV received at time=1742571423 on cpu 15 ***                 
PC: @     0x559e755018d0  (unknown)  dfly::detail::ascii_unpack_simd()

This was a replica, that I had executed below command on. I was trying to understand what data DF reports back when BGSAVEs are in progress.

redis-cli bgsave; sleep 1; redis-cli info persistence

Second occurrence

I20250321 16:48:46.963125    13 dflycmd.cc:345] Started sync with replica 100.66.130.134:6379
*** SIGSEGV received at time=1742575729 on cpu 10 ***                                        
PC: @     0x55cbc6e148d0  (unknown)  dfly::detail::ascii_unpack_simd()            

This was a recently promoted master, that looks to have started full resyncs with replicas.

To Reproduce
N/A

Expected behavior
N/A

Screenshots
N/A

Environment (please complete the following information):

• OS: Using docker.dragonflydb.io/dragonflydb/dragonfly:v1.28.0
• Kernel: # Command: uname -a Linux redis-test-0-server-0 5.15.113-flatcar #1 SMP Fri Jun 16 16:42:41 -00 2023 x86_64 x86_64 x86_64 GNU/Linux
• Containerized?: Kubernetes
• Dragonfly Version: 1.28.0

Reproducible Code Snippet

Unable to reproduce

Additional context

Let me know if there's anything else to show.

[romange]

@jdheyburn can you please run info all on the datastore - to show what types/commands are being used during your tests?

[romange]

also, are there any special arguments that you pass to dragonfly?

[jdheyburn]

We used Docker image docker.dragonflydb.io/dragonflydb/dragonfly:v1.27.3, passing in args --flagfile=/data/conf/redis.conf`.

The contents of that file were:

--dir=/data
--port=6379
--alsologtostderr
--dbfilename=dump
--df_snapshot_format=true
--info_replication_valkey_compatible
--proactor_threads=4
--replica_priority=50
--replication_acks_interval=1000

The instance was ephemeral for testing, it's no longer available unfortunately for me to obtain the info all command.