All of these services start up goroutines and then wait for the
goroutines to finish. Each of them has a `tsdb.PointBatcher` that may
return a point during the shutdown sequence. During the shutdown
sequence, a lock was held. This lock may get accessed when attempting to
write the point that came back from the `tsdb.PointBatcher`. This caused
the read lock attempt to wait forever for the write lock to be unlocked
during `Close()`.
This modifies these methods so that the write lock is released while
waiting for goroutines to finish in these three services.
If multiple tombstones exists for a series that ended up causing the
full data to be deleted, the blocks were not removed from the offsets
in the index. This causes the TSMReader to report that a key exist
but does not have any data.
During a compaction, every key should have at least one value. Since
this invariant was broken, the compaction aborted early and ends up
dropping all series keys that are lexigraphically greater than where
the breakage occured. This would cause data to be dropped during the
compaction.
This fixes a potential bug where the BlockIterator would skip blocks
if the underlying TSMReader had deletes on it concurrently. This
could possibly occur due to changes in 91eb9de3 that now use the
existing TSMReaders from the FileStore instead of creating new ones
during compaction.
TSI did not check that the max select series limit during planning
the same way that inmem did. This means that the limit could be
set but the planning of a high cardinality query would still OOM
the server. This fixes that limit as well as makes the query interruptible
during planning.
There was a race on the WaitGroup where we could end up calling Add
while another goroutine was still waiting. The functions were confusing
so they have been simplified a bit since the compactions goroutines
have been reworked a lot already.
The releng scripts' tarballs had paths like `./usr/bin/influx` whereas
the build.py output had `./influxdb-1.3.5-1/usr/bin/influx`. Now the
releng scripts match the build.py output.
This adds a new flag -exact that will return exact counts instead of
estimates. The default is to return estimates since exact counting
on a problem shard could consume a lot of memory.
The scheduling logic ended up favoring more backlogged shards
too much and would starved active, less backed up shards. This
occurred because the scheduling kicks in once a second. When it
runs, it schedules as many compactions as it can. A backed up shard
would end up having more compactions to run during the loop an would
generally get to schedule them more frequently.
This now allows each shard to try and schedule one compaction at a time
which provides a more balanced approach. At some point, we'll probably
want to more directly balanc the each shards backlog vs letting it happen
somewhat randomly.
This allows the query:
SELECT mean(value) FROM cpu GROUP BY time(1d)
To function in some way that makes sense. The upper limit is implicitly
the `now()` starting time and the lower limit will be whichever interval
the lowest point falls into.
When no lower bound is specified and `max-select-buckets` is specified,
the query will only consider points that would satisfy
`max-select-buckets`. So if you have one point written in 1970, have
another point within the last minute, and then do the above query with
`max-select-buckets` being equal to 10, the older point from 1970 will
not be considered.
Jonathan A. Sternberg
Jason Wilder
Jason Wilder
Ben Johnson
Edd Robinson
Byungjin Park
Mark Rushakoff
Mark Rushakoff
lrita
Stuart Carnie