docker-untar on btrfs_graph extremely slow

[ramelito]

Environment:

• cf v214
• diego 0.1398.0
• cf cli v6

Reproducing:
cf docker-push redis altoros/redis

on the cell vm start htop and filter out with docker keyword
time to apply layers onto btrfs_graph is so long and this leads to stopping application by timeout

there are no network delays

i know there is an issue with docker and btrfs and i wonder can we switch off btrfs for garden-linux or are there any knobs to tune it out?

[cf-gitbot]

We have created an issue in Pivotal Tracker to manage this. You can view the current status of your issue at: https://www.pivotaltracker.com/story/show/102564552.

[emalm]

Hi, @ramelito,

It sounds like this is more of an issue with the operation of garden-linux, rather than with Diego itself, so it would be better to raise this as an issue on the garden-linux repo or its BOSH release. At this point garden-linux is committed strongly to btrfs as the preferred filesystem for its graph layers, although the Garden team can comment more comprehensively on their decision. In our experience on the Diego team, it has been slow in some regards, but generally we haven't seen performance problems with it on that magnitude, so it's possible that something unexpected is going wrong.

Thanks,
Eric, CF Runtime Diego PM

[emalm]

Closing. As stated earlier, if you're still seeing this issue, please raise it with the garden-linux project for better visibility with the CF Garden team. It's also worth noting that we've observed some other issues running btrfs under heavy workloads, and the Garden team is working on alternate approaches for managing container layers that still meet CF's requirements. Please consult their team tracker for the current state of work.

Thanks again,
Eric