There is a strange race condition where a query can be killed and finish
at approximately the same time. If this happens, the query gets
retrieved by the killing task, the query gets closed by the normal
processing thread, and then the killing task attempts to kill the query
afterwards. Since the close doesn't mark the query as already killed
(since it's not killed, just merely unused), the killing thread attempts
to close the channel again.
Mark the query as killed whenever it is closed to prevent a double close
from happening. This should never cause the status to be erroneously
reported since the query status is removed from the query table within
the same lock scope.
When refactoring the query engine, I thought calling
`count(distinct(value))` multiple times was disallowed and so the
refactor made it so that wasn't possible.
It turns out that this pattern is allowed because since the distinct is
nested, it is aggregated anyway and can be combined with other
aggregates.
This removes the erroneously placed restriction.
If the close happens when next is being called, it can result in a race
condition where the current iterator gets set to nil after the initial
check.
This also fixes the finalizer so it runs the close method in a goroutine
instead of running it by itself. This is because all finalizers run on
the same goroutine so a close that takes a long time can cause a backup
for all finalizers. This also removes the redundant call to
`runtime.SetFinalizer` from the finalizer itself because a finalizer,
when called, has already cleared itself.
The previous sha was taken from a revision on a devel branch that I
thought would continue staying in the tree after it was merged. That
revision was rebased away and the API was changed for the logger.
This updates the usage of the logger and adds a simple package for
constructing the base logger.
The 1.0 version of zap changed the format of the default console logger
so this change moves over to this new logger instead of attempting to
retain backwards compatibility with the old format.
This commit adds time support to SHOW TAG VALUES. Time can be used as
both a lower and upper boundary. However, there are some caveats.
For the `inmem` index, filtering by time will still return all results
because the index data is shared across shards.
For the `tsi1` index, filtering by time will only work down to the shard
lever. Specifically, when querying by time all shards within that time
range will be used to generate the results.
When a meta query does not include a time component then it can be
answered exclusively by the index. This should result in a much faster
query execution that if the TSM engine was engaged.
This commit rewrites the following queries such that they make use
of the index where no time component is present:
- SHOW MEASUREMENTS
- SHOW SERIES
- SHOW TAG KEYS
- SHOW FIELD KEYS
This prevents a channel from being closed twice when the task manager is
closed. That same check existed to make sure a panic didn't happen when
detaching a killed query, but the check was forgotten when closing the
task manager.
This also adds a new error when attempting to kill an already killed
query.
* Introduces EXPLAIN ANALYZE command, which
produces a detailed tree of operations used to
execute the query.
introduce context.Context to APIs
metrics package
* create groups of named measurements
* safe for concurrent access
tracing package
EXPLAIN ANALYZE implementation for OSS
Serialize EXPLAIN ANALYZE traces from remote nodes
use context.Background for tests
group with other stdlib packages
additional documentation and remove unused API
use influxdb/pkg/testing/assert
remove testify reference
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.
Field math works similar to condition evaluation, but not the exact same
because we have more information to work with in field expressions than
we do in conditional math because fields retain the information about
their source while conditions do not.
The main difference is that you cannot add an unsigned literal to the
output of an integer iterator while you can inside of a condition. You
can perform math on a positive integer literal to an unsigned iterator.
Inside of the condition, we aren't sure if an integer is because of a
literal or because of an iterator so we can't make that distinction.
This more accurately shows whether or not a query has been killed.
Instead of automatically removing it from the query table when it's
killed even though goroutines and iterators may still be open, it now
marks the process as killed. This should allow us to more accurately
determine if a query has been stalled and is still using resources on
the server.
This is related to #8848, but not directly connected.
Originally, casting was performed inside of the query engine especially
for call iterators. Currently, the engine takes care of all casting so
we just need to normalize the iterators types for type safety reasons
rather than actual functional reasons.
Removing this code. Cover coverage showed that it was not hit when run
against the actual server. I ran the tests package and got code coverage
of the query package while running the tests in that package.
All time ranges are combined with AND regardless of context and
regardless of whether it makes any logical sense.
That was the previous behavior and, unfortunately, a lot of people rely
on it.
It prints the statistics of each iterator that will access the storage
engine. For each access of the storage engine, it will print the number
of shards that will potentially be accessed, the number of files that
may be accessed, the number of series that will be created, the number
of blocks, and the size of those blocks.
There are several places in the code where comma-ok map retrieval was
being used poorly. Some were benign, like checking existence before
issuing an unconditional delete with no cleanup. Others were potentially
far more serious: assuming that if 'ok' was true, then the resulting
pointer retrieved from the map would be non-nil. `nil` is a perfectly
valid value to store in a map of pointers, and the comma-ok syntax is
meant for when membership is distinct from having a non-zero value.
There was only one or two cases that I saw that being used correctly for
maps of pointers.
Previously, subqueries would honor their own ordering. We never really
supported that and I have no idea if it would work since most parts in
the query engine assume that points are being delivered in only one
ordering.
Subqueries have now been modified so if a person tries to do different
ordering, they get an error when running the query. If they specify an
ordering in the top most query, that ordering gets propagated to all
subqueries.
Fixes#8699.
Now, the prepared statement keeps the open resource and closing the open
resource created from `Prepare` is the responsibility of the prepared
statement.
This also nils out the local shard mapping after it is closed to prevent
it from being used after it is closed.
This refactors the validation code so it is more flexible and performs a
small bit of work to make preparing and executing the query easier.
The general idea is that compilation will eventually do more heavy
lifting in creating the initial plan and prepare will construct an
actual plan rather than just doing some basic field rewriting.
This change at least sets us up for that change in the future and moves
the validation code to the query execution instead of in the parser.
This also frees up the parser to parse the complete AST without worrying
if the query itself is valid. That could be useful for client code that
wants to compile a partial query to an AST and then perform
modifications on the AST for some reason.
The first call is to compile the query. This performs some initial
processing that can be done before having any access to the shards. At
the moment, it does very little, but it's intended to be changed to
eventually perform initial validations of the query and create an
internal graph structure for the execution of the query.
The second call is to prepare the query. This step has access to the
shard mapper. Right now, it just maps the shards and rewrites the fields
of the query for any wildcards. In the future, it is intended to do the
above, but also to prepare the final directed acyclical graph that will
execute the query.
The third call is to select the query. This step is intended to create
all of the iterators for processing the query. At the moment, much of
the work intended for the second step is performed in the third step.
When merging streams of system iterators we don't use tags or time.
Instead we add series keys (in the case of, for example, `SHOW SERIES`)
to the `Aux` field of the iterators' elements. This is because we only
emit merged and sorted sets of series key to the client.
We currently use `SortedMergeHeap`s to merge together multiple
iterators, and the comparitor function did not consider `Aux` fields
when determining which heap to pop the next item off during a merge. As
such, `SHOW SERIES` and `SHOW TAG KEYS` (any meta query that gets
converted into a special type of `SELECT`) were returning results in
arbitrary order.
This issue was never noticed on the `inmem` index because the streams
are always duplicates of each other, and of course it doesn't matter if
you arbitrarily merge together two idential, sorted streams...
The issue first manifested itself on the `tsi1` index, but this fix will
apply to both indexes.
The statement rewriting logic should be in the query engine as part of
preparing a query. This creates a shard mapper interface that the query
engine expects and then passes it to the query engine instead of
requiring the query to be preprocessed before being input into the query
engine. This interface is (mostly) the same as the old interface, just
moved to a different package.
This will allow refactoring the query engine and the select statement
more easily since fewer code locations will need to be changed. It also
reduces the amount of code while still keeping individual tests that are
filterable.
The ConditionExpr function is more accurate because it parses the
condition and ensures that time conditions are actually used correctly.
That means that attempting to combine conditions with OR will not result
in the query silently pretending it's an AND and nested conditions work
correctly so there is only one way to read the query.
It also extracts the non-time conditions into a separate condition so we
can stop attempting to parse around the time conditions in lower layers
of the storage engine. This change does not remove those hacks, but a
following commit should be able to sanitize the condition and remove
them.
This change provides a clear separation between the query engine
mechanics and the query language so that the language can be parsed and
dealt with separate from the query engine itself.
Jonathan A. Sternberg
Edd Robinson
Ben Johnson
Stuart Carnie
Joe LeGasse