Evicting on small disks

[2a46m4]

Evicting on very small disks will produce an error where the reverse map entry is not found. This shouldn't occur, so I added a panic to find the cause.

panic!("Couldn't find chunk {:?} in {:#?}", chunk, bucket_guard);

This produces a panic below:

thread '<unnamed>' panicked at oxcache/src/cache/mod.rs:383:21:
Couldn't find chunk ChunkLocation { zone: 1, index: 1 } in BucketMap

Reproduction

Disk values:

sector_size=4096
zone_size=128
zone_capacity=0
disk_size=1 # Gigabyte
max_open=3
max_active=3
zasl=5
img=""
hostdir=""
cores=$(nproc)

My config file.

log_level = "debug"

[server]
socket = "/tmp/oxcache.sock"
disk = "/dev/nvme0n1"
writer_threads = 3
reader_threads = 3
chunk_size = 67108864 # 64MB
block_zone_capacity = 1129316352
max_write_size = 131072

[remote]
remote_type = "emulated" # emulated | S3
bucket = "S3_BUCKET"
remote_artificial_delay_microsec = 40632

[eviction]
eviction_policy = "chunk"
# These numbers are in chunks
high_water_evict = 10 # Number remaining from end, evicts if reaches here
low_water_evict = 15  # Evict until below mark
high_water_clean = 4 # Number of invalids when we begin cleaning
low_water_clean = 2  # Clean until below mark (only used with chunk)
eviction_interval = 1  # Evict every 1s

[metrics]

I run the ./rsync_and_compile.sh script to produce the error.

Potential cause

We update the eviction LRU before updating the map, there is a chance (given an appropriate eviction threshold) that the LRU will pick it as a eviction target before the map is updated. Then the reverse mapping will return None instead of Some.