redpanda
redpanda copied to clipboard
chunk-starvation
Depending on a particular workload there might be a situation in which the chunk appender, might take on full 32MB and wait to synchronize all 32MB in 16KB chunks.
strategy to improve
- for the quota reservation of the memory, we should use 128KB
- 16KB for follow up chunks
- eager fallocation without synchronization (waiting for all chunks to finish
cc: @rkruze @mmaslankaprv @Lazin