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.
The previous default was just to have no limit at all. This adds a
configuration option to the planner so a static value can be set for the
memory limit on each individual query.
The logger is now threaded into the query controller, executor, and the
dispatcher so that we can log panics. They are logged at the info level
because the panics do not result in the system crashing and becoming
unusable.
The call to `setErr` would grab a lock that `Pop` used, but `setErr`
requires the controller run loop to be executing for it to work. If we
reverse the order of these calls, it should be fine.
When the controller moves to one of the finished states, it will finish
the parent span so that can be recorded. It presently will do this
multiple times when transitioning between different finished states.
This normally happens within the finishing states, but when compiling or
queueing fails it never enters those finished states and is instead
discarded. We need to signal that the query itself has finished in the
metrics.
This documents the responsibilities of what the Controller does and is
expected to do. It describes some behaviors that aren't implemented, but
acts as a guide for what the Controller should do as we continue
developing the query engine and improving the internal mechanics.
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.
If a query is attempting to be enqueued and it gets canceled, it will
now stop attempting to add it to the new queries queue and return the
error reported by the context. This allows the http server to cancel a
running query when the client disconnects for whatever reason without
continuing to attempt to process the canceled query.
Introduces the Statisticser interface which ResultIterators may
implement.
The HTTP implementation uses HTTP trailers to preserve the statistics.
This way we do not need to have all encoders and decoders support
statistics.
It was effectively a copied and pasted platform.ID, so change it to a
type alias. Once our known references to the query/id package are
updated to platform.ID, we'll delete the package.
Christopher M. Wolff
Mark Rushakoff
Nathaniel Cook
Stuart Carnie
Jonathan A. Sternberg
Michael Desa
Leonardo Di Donato
jlapacik