Repository navigation
seaweedfs: volumeSizeLimitMB has no per-instance override while volume.size is unbounded #4248
Description
packages/system/seaweedfs/values.yaml sets master.volumeSizeLimitMB for every tenant instance, and packages/extra/seaweedfs/values.yaml exposes no way to change it. volume.size is per-tenant and has no upper bound, so the two drift apart the moment anyone provisions a large PVC.
Those two numbers are halves of the same thing: a volume server started with -max 0 gets its slot count from free disk divided by the master's limit. A tenant on 10Gi and a tenant on 5Ti get identical volume sizes and wildly different volume counts. At 250MB a 5Ti server carries roughly 21k volumes, each a .dat and an .idx.
Deriving the limit from volume.size is not a straight substitution. Pools and zones each carry their own PVC size while volumeSizeLimitMB is master-wide, so a derived value would have to track the smallest pool. That is the tradeoff to settle, not a reason the knob should not exist.
Activity
lexfrei commentedon Sep 14, 2026
Tracking the smallest pool is not a reason to skip the knob. It is the correct semantics for a master-wide divisor: one limit has to be safe for every pool, and the smallest pool is what makes it safe.
The tradeoff to weigh is what that costs the largest pool. A 10Ti pool inheriting a limit derived from a 10Gi neighbour carries around 42k volumes, where a limit sized for its own disk would leave it a few hundred. So the choice is between a derivation that is correct and lopsided, and a flat value that is explicit and right for one disk size only.
fix(seaweedfs): raise the bucket ceiling by sizing volumes smaller (c…