Re: [Gluster-devel] Barrier design issues wrt volume snapshot

[Krishnan Parthasarathi]

To summarise the proposed solutions of the open issues,

1) For changelog xlator to have snapshot consistent, consumable changelogs (ie., those that are rotated), it needs to

    block unlink/rename/rmdir FOPs in their call path, drain in-flight operations and ensure that the to be rotated changelog

    has the corresponding changelog entries, before acknowledging to glusterd that the brick is ready to be snapshotted.

2) The solution for the inconsistency problem in a pure distribute problem is to be solved at server xlator's resolver code,

    where dentry operations on a given dentry are serialized. This is to be solved independent of barrier xlator.

Feel free to add things that the summary fails to capture.

thanks,

Krish

---

---

From: "Anand Avati" <address@hidden>
To: "Vijay Bellur" <address@hidden>
Cc: "Krishnan Parthasarathi" <address@hidden>, "Anand Avati" <address@hidden>, "Raghavendra Gowdappa" <address@hidden>, "Varun Shastry" <address@hidden>, "Pranith Kumar Karampuri" <address@hidden>, "Venky Shankar" <address@hidden>, "Kaushal M" <address@hidden>, "Rajesh Joseph" <address@hidden>, "Kotresh Hiremath Ravishankar" <address@hidden>, address@hidden
Sent: Friday, March 7, 2014 12:21:54 AM
Subject: Re: [Gluster-devel] Barrier design issues wrt volume snapshot

On Thu, Mar 6, 2014 at 12:21 AM, Vijay Bellur <address@hidden> wrote:

Adding gluster-devel.

On 03/06/2014 01:15 PM, Krishnan Parthasarathi wrote:

All,

In recent discussions around design (and implementation) of the barrier
feature, couple of things came to light.

1) changelog xlator needs barrier xlator to block unlink and rename FOPs
    in the call path. This is apart from the current list of FOPs that are blocked
    in their call back path.
    This is to make sure that the changelog has a bounded queue of unlink and rename FOPs,
    from the time barriering is enabled, to be drained, committed to changelog file and published.

Why is this necessary?

FOPs that are still coming through after enabling barrier (assuming that barrier is done in the call path) would end up in a non-consumable changelog. For these operations, geo-rep would resort to FS crawl based on xtime which does not handle unlinks and renames.

2) It is possible in a pure distribute volume that the following sequence of FOPs could result
    in snapshots of bricks disagreeing on inode type for a file or directory.

    t1: snap b1
    t2: unlink /a
    t3: mkdir /a
    t4: snap b2

where, b1 and b2 are bricks of a pure distribute volume V.

The above sequence can happen with the current barrier xlator design, since we allow unlink FOPs
to go through to the disk and only block their acknowledgement to the application. This implies
a concurrent mkdir on the same name could succeed, since DHT doesn't serialize unlink and mkdir FOPs,
unlike AFR.

Avati,

I hear that you have a solution for problem 2). Could you please start the discussion on this thread?
It would help us to decide how to go about with the barrier xlator implementation.

The solution is really a long pending implementation of dentry serialization in the resolver of protocol server. Today we allow multiple FOPs to happen in parallel which modify the same dentry. This results in hairy races (including non atomicity of rename) and has been kept open for a while now. Implementing the dentry serialization in the resolver will "solve" 2 as a side effect. Hence that is a better approach than making changes in the barrier translator.

Avati