commit 0a516732472cb233993dc61f93f7be25e483872f
Author: Chiyoung Seo <[email protected]>
Commit: Steve Yen <[email protected]>

    MB-3255 - Avoid the hashtable visit in the vbucket deletion.

    We observed that the vbucket deletion from memory hashtable
    sometimes takes more than 5 sec on Windows, and the hashtable
    visit to update the memory usage consumes most of its execution
    time. This change fixes the issue by maintaining the aggregated
    memory usage by all the items in each hashtable.

    Change-Id: I425504a23b3c018a2d10f4d3913b21f0f66f0d47
    Reviewed-on: http://review.membase.org/4251
    Reviewed-by: Dustin Sallings <[email protected]>
    Tested-by: Dustin Sallings <[email protected]>
    Tested-by: Chiyoung Seo <[email protected]>
    Reviewed-by: Steve Yen <[email protected]>

commit 650688c5976719befdc97ccc50ce3479da98e563
Author: Chiyoung Seo <[email protected]>
Commit: Dustin Sallings <[email protected]>

    MB-3246 - Destroy all sqlite statement instances in DB close function.

    DB close function fails when flush_all is executed because sqlite
    statement instances for the meta tables are not destroyed correctly.
    This change fixes this issue by destroying all the sqlite instances
    in DB close function.

    Change-Id: Ied25ae63b89229030c4aa782992b43f4f8fbe95f
    Reviewed-on: http://review.membase.org/4255
    Tested-by: Dustin Sallings <[email protected]>
    Reviewed-by: Dustin Sallings <[email protected]>

commit ad1d02a88591411281f7366cbe76c19266376c6f
Author: Trond Norbye <[email protected]>
Commit: Dustin Sallings <[email protected]>

    Write vbucket filter in a compact form

    Instead of listing all of the individual buckets all ranges that span
    more than two numbers will be written as [start,end].

    Ex:

    { 1, 2, 3, 4, 5, 6, 7, 8, 9 } ==> { [1,9] }
    { 1, 2, 3, 4, 5, 6, 7, 8, 9, 100, 101 } ==> { [1,9], 100, 102 }
    { 1, 2, 3, 4, 5, 6, 7, 8, 9, 100, 101, 102 } ==> { [1,9], [100,102] }

    Change-Id: I6597f78f0826c339f7e155fac008cff8b7a5a79f
    Reviewed-on: http://review.membase.org/4250
    Reviewed-by: Dustin Sallings <[email protected]>
    Tested-by: Dustin Sallings <[email protected]>

commit 6d6726fb5c820f793cc7b76eaac2c9f144eacf0a
Author: Bin Cui <[email protected]>
Commit: Dustin Sallings <[email protected]>

    Mb-3259 - use 'kv_{n}' format for kv table

    After spliting one kv table into multiple tables, need to find all
    kv tables in format 'kv_{n}' and enumerate these tables to inject data.

    Change-Id: Ia51e2f0122f574ddc44099548f075f15c9786f6b
    Reviewed-on: http://review.membase.org/4236
    Reviewed-by: Dustin Sallings <[email protected]>
    Tested-by: Thuan Nguyen <[email protected]>

commit be41b630f0e3f04095cad29ee6ce57040bc6fb59
Author: Dustin Sallings <[email protected]>
Commit: Chiyoung Seo <[email protected]>

    MB-3296: Delete any expired objects at startup.

    We will load all items we find on disk, but expired items will be
    unavailable.  This would normally take care of itself, but the system
    would consume more resources than necessary for up to an hour.

    This change detects that a loaded item is expired and immediately
    deletes it as if a client had requested it.

    Change-Id: Ibe0748d8fc57903ce57dcb2b96f0093d4f98ccfa
    Reviewed-on: http://review.membase.org/4244
    Reviewed-by: Chiyoung Seo <[email protected]>
    Tested-by: Chiyoung Seo <[email protected]>

commit 81f15a4097f248e23eb7b98548d3d43f8af55f26
Author: Dustin Sallings <[email protected]>
Commit: Chiyoung Seo <[email protected]>

    Don't try to avoid loading expired items from disk.

    Due to the time that the clock initialized in memcached, this never
    actually worked, and would cause us to be unable to cleanup missing
    data (and potentially lead to duplicate keys, etc...)

    Change-Id: Ie5b058baf85244b73971ac13578e20175fd2ef11
    Reviewed-on: http://review.membase.org/4243
    Reviewed-by: Chiyoung Seo <[email protected]>
    Tested-by: Chiyoung Seo <[email protected]>

commit 947cc68a5c6bf3389f0bac4d9b27d11fb2e5a932
Author: Dustin Sallings <[email protected]>
Commit: Chiyoung Seo <[email protected]>

    Queue delete when deleting an expired object.

    Previously, a delete against an expired item would be considered
    NOT_FOUND and we wouldn't see a need to queue deletion for it.  This
    allows us to tell the difference and do the right thing.

    Change-Id: Ia97e814902039156b7f68a92723945974f67ae28
    Reviewed-on: http://review.membase.org/4242
    Reviewed-by: Chiyoung Seo <[email protected]>
    Tested-by: Chiyoung Seo <[email protected]>

commit 3055a8f68d4ee5db42ed69f4f30dce63c02fb758
Author: Trond Norbye <[email protected]>
Commit: Dustin Sallings <[email protected]>

    MB-3269 Replica is very slow after fail over and rebalance

    This was caused by using the wrong variable to determine when to
    request an ACK from the other end. The variable we used would be
    incremented every time we tried to fetch an item from our message
    queue, but the item would not be sent on the wire if the item wasn't
    in memory. This resulted in that we only checked a subset of
    the values if it was time to insert an ack...

    The test to see if the ack window was full checked the
    distance between the current sequence number and the last sequence
    number we received.

    Change-Id: Iaa2020cdceb34f24f164919e942704073923909f
    Reviewed-on: http://review.membase.org/4241
    Tested-by: Dustin Sallings <[email protected]>
    Reviewed-by: Dustin Sallings <[email protected]>

commit 8fd0e7c26f59065d7c5e8ab6a6d2e937710e9c1a
Author: Trond Norbye <[email protected]>
Commit: Dustin Sallings <[email protected]>

    Improve the stats for a tap connection

    Decode the tap connect flags and the vbucket list into a human
    readable form:

    STAT eq_tapq:anon_2:flags 20 (ack,vblist)
    STAT eq_tapq:anon_2:vb_filter { 767, 768 }

    Change-Id: I8c63246033ac1723fbe270477a899476cde3eb9a
    Reviewed-on: http://review.membase.org/4239
    Reviewed-by: Dustin Sallings <[email protected]>
    Tested-by: Dustin Sallings <[email protected]>