cluster node cannot restart, exit code 139

[uestczz]

my cluster has 3 nodes, each node has 31Gb Ram and 4cpus. I have 0.15 billion vectors of 768 dimensions. now I write most of them, but one node exit and always exits with code 139 if restarts. my data is group by groupid. at first ,I write my data all in one collection with groupid in the payload. but the search is very low and 30s timeout. so I create one collection for every groupid, just like collectionName_groupid,and there're nearly 1.5 thousand collections. my collection config is as below:
"params": { "vectors": { "size": 768, "distance": "Cosine", "on_disk": true }, "shard_number": 6, "replication_factor": 1, "write_consistency_factor": 1, "on_disk_payload": true }, "hnsw_config": { "m": 16, "ef_construct": 100, "full_scan_threshold": 10000, "max_indexing_threads": 2, "on_disk": true }, "optimizer_config": { "deleted_threshold": 0.2, "vacuum_min_vector_number": 1000, "default_segment_number": 0, "max_segment_size": null, "memmap_threshold": 10000, "indexing_threshold": 10000, "flush_interval_sec": 5, "max_optimization_threads": 1 }, "wal_config": { "wal_capacity_mb": 64, "wal_segments_ahead": 0 }, "quantization_config": null }, "payload_schema": {}

my docker startup commmand is :
docker run -itd --name qdrant03 --oom-kill-disable --memory=24g --cpus=3.5 --restart=always -p 6333:6333 -p 6334:6334 -p 6335:6335\ -v $(pwd)/path/to/data:/qdrant/storage \ -v $(pwd)/path/to/custom_config.yaml:/qdrant/config/production.yaml \ qdrant/qdrant ./qdrant --bootstrap 'http://10.0.0.49:6335'

docker logs:
Recovering collection xd_vector_service_xueqiu_1539 [00:00:00] ██ 12/12 (eta:0s)[2023-06-11T07:45:28.280Z INFO storage::content_manager::toc] Loading collection: xd_vector_service_xueqiu_405 Recovering collection xd_vector_service_xueqiu_405 [00:00:00] █ 108/108 (eta:0s)terminate called without an active exception terminate called recursively

docker inspect:
"State": { "Status": "exited", "Running": false, "Paused": false, "Restarting": false, "OOMKilled": false, "Dead": false, "Pid": 0, "ExitCode": 139, "Error": "", "StartedAt": "2023-06-11T07:16:04.029753941Z", "FinishedAt": "2023-06-11T07:45:33.660368488Z" },
it seems buffer/cache is large to 20Gi, and maybe segment default. so the memory is not enough?
so may I shoud cut off the number of my collecitons by hash routing ?

[generall]

Creating 1000+ collections is not recommended, this will create a huge overhead.
Have you followed the instructions in the https://qdrant.tech/documentation/tutorials/multiple-partitions/ ?

[generall]

Also, serving 400Gb vectors on the 32Gb machine would require a very good disks. Could you please share the IOPS measurement of the disks you are using?

[uestczz]

iops 32000
throughput 300MB/s
this measurement is provide by my coloud provider

[uestczz]

{
"name": "xd_vector_service_xueqiu",
"vectors": {
"size": 768,
"distance": "Cosine",
"on_disk": true
},
"optimizers_config": {
"memmap_threshold": 20000,
"indexing_threshold": 20000
},
"hnsw_config": {
"payload_m": 16,
"m": 0,
"on_disk": true
},
"wal_config": {
"wal_capacity_mb": 1024,
"wal_segments_ahead": 0
},
"shard_number": 6,
"replication_factor": 1,
"on_disk_payload": true
}
I prepare use this vector config to test

[uestczz]

so is my hardware satisfied? 0.15billion vector of 768d, payload format is {"id": "866_d013b877787cc04d430c87ba64a64958","sid": 2146740788, "group_id": 998}. data with different groupid is isolated. 3nodes, every node has 32G ram, 400G ssd, 4cpu. and we turn on all on_disk configurations. we try to use quantinization, but it seems slow in the first search, and we have a lot of random search for every groupid, so we don't use it again. or could you please recommend hardware measurements for my situation?