Displaying #cassandra-dev/2016-12-28.log:

Wed Dec 28 00:19:28 2016  dikang:Joined the channel
Wed Dec 28 00:21:26 2016  thorkild_:Joined the channel
Wed Dec 28 00:23:43 2016  gila_:Joined the channel
Wed Dec 28 01:33:08 2016  mstepura:Joined the channel
Wed Dec 28 02:26:01 2016  mstepura:Joined the channel
Wed Dec 28 03:26:42 2016  mstepura:Joined the channel
Wed Dec 28 03:55:12 2016  mstepura:Joined the channel
Wed Dec 28 04:15:21 2016  mstepura:Joined the channel
Wed Dec 28 04:22:10 2016  dikang:Joined the channel
Wed Dec 28 05:07:23 2016  mstepura:Joined the channel
Wed Dec 28 05:38:05 2016  cassci:Joined the channel
Wed Dec 28 05:55:52 2016  mstepura:Joined the channel
Wed Dec 28 05:57:09 2016  dikang:Joined the channel
Wed Dec 28 06:39:27 2016  mstepura:Joined the channel
Wed Dec 28 07:19:30 2016  mstepura:Joined the channel
Wed Dec 28 08:43:51 2016  kvaster:Joined the channel
Wed Dec 28 08:50:01 2016  kvaster:Joined the channel
Wed Dec 28 09:07:01 2016  iksaif: anybody familiar with both OutgoingTcpConnection, SEP and Stage.Mutation ? when debuging CASSANDRA-13039 I think I found two things
Wed Dec 28 09:07:03 2016  CassBotJr:https://issues.apache.org/jira/browse/CASSANDRA-13039 (Open; Unresolved; Unscheduled): "Mutation time mostly spent in LinkedBlockingQueue.put()"
Wed Dec 28 09:07:34 2016  iksaif:1# the current implementation creates a lot of context switches if the otc coalescing window is too small, wich then makes enqueue() super expensive/slow
Wed Dec 28 09:08:09 2016  iksaif:2# there is a potential deadlock because MUTATION messages will be executed on Stage.MUTATION and their child local mutations will be executed there too
Wed Dec 28 09:09:11 2016  iksaif:(one could say that it's fine because remote mutations could finish first, but that won't happen if the other side is in the same state)
Wed Dec 28 09:45:32 2016  iksaif:(I'm also getting a lot of corruption kind of messages when mixing the latest 3.X with ~3.11, something is fishy here)
Wed Dec 28 12:32:15 2016  gila:Joined the channel
Wed Dec 28 15:12:03 2016  ashutoshd_:Joined the channel
Wed Dec 28 15:12:12 2016  ashutoshd_:Hi All. Can someone please look at my following query: http://stackoverflow.com/questions/41364262/is-allow-filtering-in-cassandra-for-following-query-efficient
Wed Dec 28 15:26:38 2016  kvaster:Joined the channel
Wed Dec 28 16:09:33 2016  mebigfatguy:ashutoshd_, a better question for #cassandra
Wed Dec 28 16:16:47 2016  zanson:so how do we get github emails changed to commits@ instead of dev@ ?
Wed Dec 28 16:56:41 2016  thobbs:Joined the channel
Wed Dec 28 17:27:20 2016  vjarnot:Does TTL expiration ignore writetime timestamps (insert with 'using timestamp foo') on purpose, or is this a bug? Seems like the latter to me, but I don't know the history
Wed Dec 28 17:29:05 2016  thobbs:it's on purpose
Wed Dec 28 17:29:36 2016  thobbs:there's no guarantee that the write timestamp is actually milliseconds since the epoch
Wed Dec 28 17:30:01 2016  thobbs:err, micros, I mean
Wed Dec 28 17:34:45 2016  vjarnot:ugh. true, but 'contractually' that's what it's supposed to be
Wed Dec 28 17:36:43 2016  thobbs:not contractually, that's just the de facto standard
Wed Dec 28 17:36:56 2016  thobbs:especially when TTLs were introduced in 0.7
Wed Dec 28 17:39:55 2016  vjarnot:ok, but I'd argue that calling it "timestamp" implicitly defines it as exactly that, hence "contractually" in scary quotes
Wed Dec 28 17:40:06 2016  vjarnot:appreciate the reply in any case, now to find a workaround
Wed Dec 28 17:40:22 2016  thobbs:well, some people used millis since the epoch, others used micros since the epoch
Wed Dec 28 17:40:32 2016  thobbs:both timestamps
Wed Dec 28 17:41:20 2016  thobbs:out of curiosity, what are you trying to do?
Wed Dec 28 17:42:26 2016  vjarnot:both are timestamps true, but being more specific the timestamp property is defined as timestamp in micros, so it's back to the definition being one thing and the historical baggage driving behavior that differs
Wed Dec 28 17:42:39 2016  vjarnot:I'm trying to import a bunch of historical data
Wed Dec 28 17:44:26 2016  thobbs:ah, so you'll just have to calculate a different TTL for each row, based on its timestamp
Wed Dec 28 17:44:48 2016  vjarnot:yes, was hoping to avoid that
Wed Dec 28 17:53:43 2016  JayZhuang:Joined the channel
Wed Dec 28 17:54:49 2016  vjarnot:bit confusing when TWCS appears to honor the "using timestamp" timestamp, but SELECTs don't and TTL expiration doesn't
Wed Dec 28 18:03:29 2016  JayZhuang:Joined the channel
Wed Dec 28 18:27:41 2016  mstepura:Joined the channel
Wed Dec 28 19:34:28 2016  JayZhuang:Joined the channel
Wed Dec 28 19:41:59 2016  iamaleksey:vjarnot: also some people use negative values, to get first writer wins heuristics
Wed Dec 28 19:42:30 2016  iamaleksey:vjarnot: as far as Cassandra is concerned, with few exceptions like dropping columns, the timestamp is an opaque integer
Wed Dec 28 19:46:41 2016  vjarnot:thank you iamaleksey
Wed Dec 28 19:46:42 2016  vjarnot:I think at some major-version point, it'd be helpful to either change the name (from timestamp to something less timestampy) or actually treat it as a timestamp consistently rather than occasionally ... As someone approaching "using timestamp" as a means to work with TWCS, the fact that it's otherwise treated as a sequence is confusing
Wed Dec 28 19:48:47 2016  iamaleksey:treat is as a logical timestamp
Wed Dec 28 19:48:52 2016  iamaleksey:still a timestamp
Wed Dec 28 19:48:55 2016  iamaleksey:just not a unix one
Wed Dec 28 19:48:55 2016  adamholmberg:Joined the channel
Wed Dec 28 19:50:03 2016  iamaleksey:I doubt there is a better name for it; not that it'd ever change, it'd have to be Trump-level genius to justify changing something this established
Wed Dec 28 19:51:26 2016  adamholmberg:Joined the channel
Wed Dec 28 19:53:03 2016  adamholm_:Joined the channel
Wed Dec 28 19:53:34 2016  vjarnot:figured as much with regard to "something this established"
Wed Dec 28 19:54:35 2016  adamholmberg:Joined the channel
Wed Dec 28 19:54:59 2016  vjarnot:that said, "logical timestamp" ... disagree - it's a sequence; regardless of adjective applied
Wed Dec 28 19:58:04 2016  mstepura:Joined the channel
Wed Dec 28 19:58:44 2016  romainh:Joined the channel
Wed Dec 28 20:01:08 2016  vjarnot:but my gripe is not with nomenclature - it's that the timestamp is treated differently by different parts of Cassandra
Wed Dec 28 20:16:25 2016  mebigfatguy:iamaleksey, you need to make TTL great again
Wed Dec 28 20:25:54 2016  iamaleksey:mebigfatguy: (:
Wed Dec 28 20:26:17 2016  dikang:Joined the channel
Wed Dec 28 21:02:14 2016  mstepura:Joined the channel
Wed Dec 28 21:45:27 2016  romainh1:Joined the channel
Wed Dec 28 21:50:36 2016  dikang:Joined the channel
Wed Dec 28 23:50:24 2016  mstepura:Joined the channel

Comments