* task(fix): Tasks should no longer have inaccurate response data
tasks should be able to pull from a table with both success and failed results
Co-authored-by: AlirieGray <alirie@influxdata.com>
Co-authored-by: docmerlin <emrys@influxdata.com>
This replaces usages of the spec compiler with the ast compiler and it
removes the error message referencing the spec compiler as an available
input.
It does not remove any of the code using the spec compiler that is
involved for proxying requests and it does not remove it from the API.
* Update task servicetest to move dependency to the new TaskControlService
closes#12724
We will now have the capability to write new task services that dont have to implement the backend.Store or LogReader or LogWriters
Task updates now attempt to keep the existing runners working.
This causes the system to be slightly slower after a task update and caused a flakey test.
The synchronous executor was missing a call to ResultIterator.Release.
The asynchronous executor wasn't even calling Query.Statistics.
Also add a test that the scheduler records the statistics to the run
log, and that the statistics are visible from the launcher test. The
launcher test is the most likely place to catch if something goes wrong
in the full stack.
The late measurement filter, after a pivot, had the potential to result
in empty groups without a runID, which would cause a runtime error,
which would cause the whole query to fail.
Experimentation has shown that those empty tables will no longer arrive
by filtering early on measurement.
This should considerably simplify debugging when things go wrong with
the tasks, as this error can be displayed from the UI or CLI. Prior to
this change, you would have to view the console output from influxd.
Fixes#12548.
Task ID is now a required value on run and log filters. It was
effectively required by all implementations before anyway, so now those
types reflect that requirement.
Organization ID was removed from those same fields. The TaskService
looks up the organization ID via the task in cases where we need it at a
lower layer.
This was a missed case from #11817.
This case currently occurs when creating a task through the UI, using a
session rather than a full-fledged authorization. It doesn't fix that
case yet, but at least it will log an informative message.
Immediately before the executor calls out to the query service, the
executor loads the authorizer associated with the task, and associates
that authorizer with the context used to execute the query.
Accept token when creating or updating a task, but only report back the
authorization ID.
This means the executor and the platform adapter are now both aware of
an Authorization Service.
This switches run status from a tag to a field. This is likely a
breaking change to existing task logs.
Using a one-off local query, for 250 records, the previous approach took
around 10 seconds and the new approach is about 30 milliseconds. At 1000
records, the previous approach was roughly 110 seconds and the new
approach is around 70 milliseconds.
filter out resources that have mission IDs
fix(influxdb): simplify auth check in PermissionAllowed
review(platform): update as noted in review
fix(influxdb): ensure permission has valid org id
I did this with a dumb editor macro, so some comments changed too.
Also rename root package from platform to influxdb.
In interest of minimizing risk, anyone importing the root package has
now aliased it to "platform" so that no changes beyond imports were
necessary in those files.
Lastly, replace the old platform module to local path /dev/null so that
nobody can accidentally reintroduce a platform dependency while
migrating platform code to influxdb.
Also rename RetryAlreadyQueuedError by running:
gorename -from '"github.com/influxdata/platform/task/backend".RetryAlreadyQueuedError' -to RequestStillQueuedError
and some further manual cleanup for comments.
A standard Makefile is used now in all subdirs that run go generate.
Make will only generate the file if its source files changed.
The checkgenerate target runs clean to ensure all targets a generated
fresh.
Previously, the WithTicker option would call TickScheduler.Tick every
time the underlying time.Ticker sent a time on its channel. This meant
we used a 1s period, which meant that in the worst case, we would see a
tick at about 999ms after the second rollover.
This change increases the underlying time.Ticker frequency, but only
calls TickScheduler.Tick after a second rolls over. Since we now use a
tick frequency of 100ms, during normal operation, TickScheduler.Tick
will be called within 0.1s after the second rolls over.
Previously, using every=1m would run every minute from when the task was
created. This change restores the original intent, which is that "every
1m" happens on the minute, "every 1h" on the hour, etc.
The keys used to be strings, back when platform.ID was a byte slice, not
a uint64.
And rename the field from runners to meta, which is much more accurate.
The flux query controller was updated to include a Shutdown method a
while ago. Explicitly handle query controller creation and shutdown
where applicable.
In influxd, this ensures that outstanding queries are handled before the
process dies. In tests, this ensures that query controller goroutines
aren't leaked, which drastically simplifies reading full stack traces.
This change also registers query controller metrics with the prometheus
registry in influxd.
This is to ensure that Scheduler.Stop blocks until outstanding task runs
finish. There were enterprise tests failing because outstanding runs of
a task were calling (*testing.T).Log after the test finished.
Before, running go test -short -count=1 ./task/... would take about 15
seconds. By skipping these two long tests in short mode, that takes
about 10 seconds instead. In particular, the task package itself went
from 10 seconds to under a second, but there wasn't a realized 10 second
gain due to packages being tested in parallel.
feat(platform): add ToPermissions method to user resource mapping
The ToPermissions method returns a set of permissions that is granted
via a user resource mapping.
feat(bolt): resolve sessions permissions on lookup
feat(http): use authorizer instead of authorization service for write api
feat(bolt): create user resource mappings for org users in bucket create
feat(bolt): create user resource mapping for first org/user
fix(platform): use authorizer for query endpoint instead of authorization
test(http): use cmp instead of reflect for decode test
- Brought over enterprise's QueryLogReader, with small adjustments
- Time filters are for the run's ScheduledFor field, per spec
- Adjusted run log timestamps for consistent formatting:
- ScheduledFor is RFC3339 because it's a whole-second timestamp
- StartedAt, FinishedAt use RFC3339Nano for high precision
- Several test adjustments to use relative time, for easier integration
with storage retention
This also adjusts the TaskService interface's RetryRun method to accept
a task ID rather than an org ID. Internally, we still look up runs by
organization, and maybe that will change later, but this is a more
natural way for clients to look it up.
This also changes the backend.Store API to remove the EnableTask and
DisableTask methods, merging their functionality into ModifyTask, which
has been named to UpdateTask to keep closer to the CRUD acronym.
This renames TaskEnabled and TaskDisabled to TaskActive and
TaskInactive, to keep in line with the swagger and other parts of the
system. But I left the EnableTask and DisableTask methods on the Store
interface for now. I could see eliminating those methods if we adjust
the signature of the UpdateTask method.
The generate commands have been modified to take advantage of the new
functionality in Go 1.11 that allows `go run` to execute a package
instead of individual files.
This functionality combined with Go modules allows us to execute a
package directly out of our pinned dependencies rather than accidentally
picking up another binary outside of the build environment.
This also simplifies the Makefile because they no longer have to be
responsible for installing the correct tooling since the Go command
takes care of that logic. It also makes it so that the Makefiles with
file generation can now be invoked from their appropriate subdirectories
so they are contained within the directory itself rather than relying on
values in the top level Makefile.
It is now possible to generate all files within this project by using:
go generate ./...
Or the Makefile can continue to be used.
This commit also copies over the special copy of `tmpl` that the storage
engine uses within the influxdb repository. It was never copied over so
using `go generate` on these packages did not work.
We are moving the necessary code for 2.0 from the influxdb 1.X
repository to the platform 2.0 repository. The logger is an unnecessary
dependency on the old influxdb that is making life more complicated.
Introduce a RunLogBase type that encapsulates the base information
shared across multiple logs for the same run. This has the information
previously part of the API (Task and RunID), and includes the addition
of a RunScheduledFor timestamp.
Also remove the RunScheduled value for RunStatus, as it was never used
in the scheduler.
Also fix a handful of segfaults caused by improperly retained byte slices
from bolt, and combine two view transactions into one to avoid data
race when a delete happened between them.
A mistake in the WithTicker method caused it to block.
This was not evident because until recently we used a cron styler timer.
Removing the block should allow the task deamon to start up properly.
In TestScheduler_RunLog, the run log gets updated asynchronously so we
need to poll to ensure we don't read a stale value.
In TestScheduler_CreateNextRunOnTick, we were previously reading the
one entry out of two from a map, so it was often the entry we weren't
interested in.
CreateNextRun centralizes StoreTaskMeta modification and therefore
simplifies Store functions a bit.
Store.CreateNextRun isn't called anywhere yet. The next step here is to
use CreateNextRun from the scheduler and remove Store.CreateRun.
The logic used to be only in the scheduler, but now Options exposes it.
This is another step towards making StoreTaskMeta schedule-aware to
enable creating manually requested runs.
This logic was repeated in multiple places, so now it can live in a
single place. This also sets a precedent going forward, for adding
methods to the StoreTaskMeta type. We'll likely be adding more methods
to support manual runs of a task.
More specifically, introduce a `scheduleAfter` argument to
Store.CreateTask. The previous behavior was leaving a new task's meta
LatestComplete value set to zero, which meant that the first run of a
schedule would start from 1970. Now, it's set to time.Now unless
otherwise specified.
Updates #595.
We were previously doing `r.logger = r.logger.With(...)`, which ended up
duplicating the provided field in the log output. Now ensure that we
only set the run_id once in a logger.
Moves idpe.QueryService into platform/query.ProxyQueryService
Splits the Request into ProxyRequest and Request.
Changes query.QueryService and query.AsyncQueryService to use a Request
type. This means that the Compiler interface is consumed by the service
to abstract out transpilation vs Flux compilation vs raw spec.
The transpiler handler is removed.
There are separate http handlers and service implementations for each of
the three query services.
Query logging types are moved into platform.
The ResultIterator now expects Cancel to always be called.
The fluxd binary exposes the query endpoint specified in the swagger
file.
We need a coordination layer to facilitate the communication in the api to the task service.
The api was also not connected in main to the http service.
This is to work around slow parsing (#484).
This change saves about 2 seconds on tests for task/backend and
task/backend/bolt. It saves around 45 seconds on enterprise tests.
After we started parsing options inside of the task validation methods,
we accidentally shadowed the "missing script" error when the script was
empty. This restores the behavior of telling the user "missing script"
rather than a cryptic syntax error.
Lyon Hill
Alirie Gray
Nathaniel Cook
Christopher M. Wolff
Delmer
Todd Persen
Jonathan A. Sternberg
Jacob Marble
kelwang
zhulongcheng
Jeff Wendling
docmerlin (j. Emrys Landivar)
Mark Rushakoff
Chris Goller
Jacob Marble
Michael Desa
Lorenzo Affetti
Mark Rushakoff
Andrew Watkins
Mark Rushakoff
Edd Robinson
Ben Johnson
jlapacik
Adam
Leonardo Di Donato
Brandon Farmer
Stuart Carnie
Jade McGough