Project

General

Profile

Bug #10894

bpobj_enqueue_subobj() should copy small subobj's

Added by Jerry Jelinek about 1 year ago.

Status:
New
Priority:
Normal
Assignee:
Category:
zfs - Zettabyte File System
Start date:
2019-05-01
Due date:
% Done:

0%

Estimated time:
Difficulty:
Medium
Tags:
needs-triage
Gerrit CR:

Description

ZoL: 9553c533a63 bpobj_enqueue_subobj() should copy small subobj's

When we delete a snapshot, we consolidate some bpobj's together because
we no longer need to keep their entries in separate buckets. This is
done in constant time by including the "sub" bpobj by reference in the
parent bpobj.

After many snapshots have been deleted, we may have many sub-bpobj's.
Usually, most sub-bpobj's don't contain many BP's. Compared to this
small payload, the sub-bpobj is relatively heavyweight since it is a
object in the MOS. A common scenario on a long-lived pool is for the
vast majority of MOS objects to be small sub-bpobj's.

To improve this situation, when consolidating bpobj's together,
bpobj_enqueue_subobj() can copy the contents of small bpobj's into the
parent, and then delete the enqueued bpobj, rather than including it by
reference. Since this copying is limited in size (to one block), the
consolidation is still constant time, though with a larger constant due
to reading in the one block of the enqueued bpobj.

This idea and mechanism are similar to how we handle "sub-subobj's".
When including a sub-bpobj by reference, if the sub-bpobj itself has
less than a block of sub-sub-bpobj's, the list of sub-sub-bpobj's is
copied to the parent bpobj's list of sub-bpobj's.

Also available in: Atom PDF