[
https://issues.apache.org/jira/browse/CASSANDRA-6602?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14172990#comment-14172990
]
Björn Hegerfors commented on CASSANDRA-6602:
--------------------------------------------
[~krummas] Yep, I also think of seconds/minutes/etc. when I hear 'time unit'. I
agree time_unit is not the best name for what it ended up being. I guess I've
tested time_unit with values like 60,000,000 (minute) and 3,600,000,000 (hour),
and then timeUnit made sense in the code, when setting the initial target size
to 1 of whatever the 'time unit' is (e.g. 1 hour). But if you use something
like 300,000,000 (5 minutes), calling that a 'time unit' is a bit iffy. It
feels like 'minute' is the time unit, and 5 some multiplier on how big the
initial target is (so 5 minute units, rather than one 5-minute unit).
Then you noted the other problem, that can cause even more confusion. The
potentially differing timestamp formats is probably a harder nut to crack, put
in relation with the naming. But the solution to it probably also affects the
number of options and their names. You could even go with three options
replacing time_unit: timestamp_resolution, time_unit, base_time. Then, for 5
minutes with microsecond timestamps, you would specify (with longs or strings,
I don't know) timestamp_resolution=1,000,000 (microseconds), time_unit=60
(minutes), base_time=5. The time_unit that my code uses is simply the product
of these options.
What remains, as far as I can see, is a 2-option solution: remove the middle
option (time_unit) and default to minutes or seconds (I personally think
seconds are nicer). Or even microseconds, but then base_time (or
base_time_microseconds) has to be a big number and the product should be
timestamp_resolution * base_time / 1,000,000. The latter choice is only there
if someone believes that sub-second targets could be useful.
Regardless, I think it's important to make it clear to all users that they have
to make sure that the 'timestamp resolution' is correct. One way would be to
default to microseconds and simply put a visible warning in the documentation
about DTCS that it expects microseconds, so if you use something else, you need
to change the timestamp_resolution option. I don't know if Cassandra somehow
prefers microseconds or if it likes to stay neutral and not estrange those who
use something else. But CQL has that default, doesn't it? Another way to
prevent bugs caused by non microsecond timestamps would be to simply require a
timestamp_resolution to be defined. No defaults.
I don't know if I've helped narrowing anything down here, but these are all the
alternatives that I can think of. Without knowing what conventions there might
be to things like this, my preferred choice right now is probably these
options: long timestamp_resolution (default: 1,000,000), long base_time_seconds
(default: 3600? 300?). That and the warning in the documentation, somewhere
visible. My time to ask: wdyt?
> Compaction improvements to optimize time series data
> ----------------------------------------------------
>
> Key: CASSANDRA-6602
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6602
> Project: Cassandra
> Issue Type: New Feature
> Components: Core
> Reporter: Tupshin Harper
> Assignee: Björn Hegerfors
> Labels: compaction, performance
> Fix For: 2.0.11
>
> Attachments: 1 week.txt, 8 weeks.txt, STCS 16 hours.txt,
> TimestampViewer.java,
> cassandra-2.0-CASSANDRA-6602-DateTieredCompactionStrategy.txt,
> cassandra-2.0-CASSANDRA-6602-DateTieredCompactionStrategy_v2.txt,
> cassandra-2.0-CASSANDRA-6602-DateTieredCompactionStrategy_v3.txt
>
>
> There are some unique characteristics of many/most time series use cases that
> both provide challenges, as well as provide unique opportunities for
> optimizations.
> One of the major challenges is in compaction. The existing compaction
> strategies will tend to re-compact data on disk at least a few times over the
> lifespan of each data point, greatly increasing the cpu and IO costs of that
> write.
> Compaction exists to
> 1) ensure that there aren't too many files on disk
> 2) ensure that data that should be contiguous (part of the same partition) is
> laid out contiguously
> 3) deleting data due to ttls or tombstones
> The special characteristics of time series data allow us to optimize away all
> three.
> Time series data
> 1) tends to be delivered in time order, with relatively constrained exceptions
> 2) often has a pre-determined and fixed expiration date
> 3) Never gets deleted prior to TTL
> 4) Has relatively predictable ingestion rates
> Note that I filed CASSANDRA-5561 and this ticket potentially replaces or
> lowers the need for it. In that ticket, jbellis reasonably asks, how that
> compaction strategy is better than disabling compaction.
> Taking that to heart, here is a compaction-strategy-less approach that could
> be extremely efficient for time-series use cases that follow the above
> pattern.
> (For context, I'm thinking of an example use case involving lots of streams
> of time-series data with a 5GB per day ingestion rate, and a 1000 day
> retention with TTL, resulting in an eventual steady state of 5TB per node)
> 1) You have an extremely large memtable (preferably off heap, if/when doable)
> for the table, and that memtable is sized to be able to hold a lengthy window
> of time. A typical period might be one day. At the end of that period, you
> flush the contents of the memtable to an sstable and move to the next one.
> This is basically identical to current behaviour, but with thresholds
> adjusted so that you can ensure flushing at predictable intervals. (Open
> question is whether predictable intervals is actually necessary, or whether
> just waiting until the huge memtable is nearly full is sufficient)
> 2) Combine the behaviour with CASSANDRA-5228 so that sstables will be
> efficiently dropped once all of the columns have. (Another side note, it
> might be valuable to have a modified version of CASSANDRA-3974 that doesn't
> bother storing per-column TTL since it is required that all columns have the
> same TTL)
> 3) Be able to mark column families as read/write only (no explicit deletes),
> so no tombstones.
> 4) Optionally add back an additional type of delete that would delete all
> data earlier than a particular timestamp, resulting in immediate dropping of
> obsoleted sstables.
> The result is that for in-order delivered data, Every cell will be laid out
> optimally on disk on the first pass, and over the course of 1000 days and 5TB
> of data, there will "only" be 1000 5GB sstables, so the number of filehandles
> will be reasonable.
> For exceptions (out-of-order delivery), most cases will be caught by the
> extended (24 hour+) memtable flush times and merged correctly automatically.
> For those that were slightly askew at flush time, or were delivered so far
> out of order that they go in the wrong sstable, there is relatively low
> overhead to reading from two sstables for a time slice, instead of one, and
> that overhead would be incurred relatively rarely unless out-of-order
> delivery was the common case, in which case, this strategy should not be used.
> Another possible optimization to address out-of-order would be to maintain
> more than one time-centric memtables in memory at a time (e.g. two 12 hour
> ones), and then you always insert into whichever one of the two "owns" the
> appropriate range of time. By delaying flushing the ahead one until we are
> ready to roll writes over to a third one, we are able to avoid any
> fragmentation as long as all deliveries come in no more than 12 hours late
> (in this example, presumably tunable).
> Anything that triggers compactions will have to be looked at, since there
> won't be any. The one concern I have is the ramificaiton of repair.
> Initially, at least, I think it would be acceptable to just write one sstable
> per repair and not bother trying to merge it with other sstables.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
