mirror of
https://github.com/seaweedfs/seaweedfs.git
synced 2026-08-25 16:47:28 +00:00
Chris Lu
bd3b75b534
The inode table only shrinks when the kernel returns a reference, so a walk over a large tree grows it with every child it touches and nothing gives any of it back until the kernel reclaims dentries. On a mount with tens of millions of files that is most of the process's heap. -maxInodeEntries stops the speculative half of that. Past the cap a readdirplus still reports every name, but leaves the EntryOut zeroed: Linux reads a zero nodeid as 'no attributes for this entry', takes no reference, and looks up the ones the client actually needs. A LOOKUP is never refused, so the table can still grow past the cap by what a client really asks for, and nothing here evicts. Default 0 keeps today's behavior.