EC Object 404 is too slow

Bug #1663742 reported by clayg
12
This bug affects 2 people
Affects Status Importance Assigned to Milestone
OpenStack Object Storage (swift)
Confirmed
Medium
Unassigned

Bug Description

We tend to dig 2 * replicas into handoffs before we return 404 - on big EC policies (e.g. 15+3) that's a lot of requests.

X-Newest: true on a HEAD request will also dig super deep

Revision history for this message
Samuel Merritt (torgomatic) wrote :

Yeah, this should probably be configurable separately.

What's a good default value to use? 2 * number-of-parity-fragments?

Changed in swift:
status: New → Confirmed
importance: Undecided → Medium
Revision history for this message
clayg (clay-gerrard) wrote :

no, I think on EC if you hit every primary and find *nothing* you can 404 - we probably don't even need to hit EVERY primary in a 15+3 schema.

But the question is more complex if you found 12 frags by the time you exhaust primaries... like you can probably guess the rest are out there... somewhere...

Ideally we'd respond with the data when we can.

The question is when can we say "yup, probably don't have this"

Replicas used to also not dig past primaries - then one day someone was like "yeah all three replicas were on handoffs after PUT because of 507's" and we were like, ok better dig!

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.