cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benedict (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-6630) Replace UnsortedColumns with ArrayBackedSortedColumns
Date Thu, 30 Jan 2014 16:30:14 GMT


Benedict commented on CASSANDRA-6630:

bq. Why not, though I'll note this could easily back-fire performance wise if some accesses
are intermingled with adds

I don't see how it would end up worse. If you track how far you've sorted the list to, you
just sort the end unsorted portion, and then merge it into the sorted portion. Worst case
behaviour is the same as currently, algorithmically. Might need to do something clever to
avoid allocating extra memory when merging (like, we might have to manually shift instead
of relying on ArrayList), but nothing too dramatic.

bq. "severely pathological" sounds a bit like over-dramatization

Well, let's say "pathological". If you have a thousand updates, it will require a 500k ops,
which may be on the order of 100 cycles each. This is a latency of 25ms just for this portion.
Let's say somebody creates a statement with 10k columns (not crazy, but not common); you're
now talking 2.5s. This is pathological in my book.

> Replace UnsortedColumns with ArrayBackedSortedColumns
> -----------------------------------------------------
>                 Key: CASSANDRA-6630
>                 URL:
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Aleksey Yeschenko
>            Assignee: Aleksey Yeschenko
>             Fix For: 2.1
>         Attachments: 6630.txt
> It's possible for a counter mutation to have several CounterUpdateCell-s with the same
cell name. Those should be summed up and not ignored.

This message was sent by Atlassian JIRA

View raw message