# ⚠️ Breaking
Many deprecated queue config options are removed (actually, they should
have been removed in 1.18/1.19).
If you see the fatal message when starting Gitea: "Please update your
app.ini to remove deprecated config options", please follow the error
messages to remove these options from your app.ini.
Example:
```
2023/05/06 19:39:22 [E] Removed queue option: `[indexer].ISSUE_INDEXER_QUEUE_TYPE`. Use new options in `[queue.issue_indexer]`
2023/05/06 19:39:22 [E] Removed queue option: `[indexer].UPDATE_BUFFER_LEN`. Use new options in `[queue.issue_indexer]`
2023/05/06 19:39:22 [F] Please update your app.ini to remove deprecated config options
```
Many options in `[queue]` are are dropped, including:
`WRAP_IF_NECESSARY`, `MAX_ATTEMPTS`, `TIMEOUT`, `WORKERS`,
`BLOCK_TIMEOUT`, `BOOST_TIMEOUT`, `BOOST_WORKERS`, they can be removed
from app.ini.
# The problem
The old queue package has some legacy problems:
* complexity: I doubt few people could tell how it works.
* maintainability: Too many channels and mutex/cond are mixed together,
too many different structs/interfaces depends each other.
* stability: due to the complexity & maintainability, sometimes there
are strange bugs and difficult to debug, and some code doesn't have test
(indeed some code is difficult to test because a lot of things are mixed
together).
* general applicability: although it is called "queue", its behavior is
not a well-known queue.
* scalability: it doesn't seem easy to make it work with a cluster
without breaking its behaviors.
It came from some very old code to "avoid breaking", however, its
technical debt is too heavy now. It's a good time to introduce a better
"queue" package.
# The new queue package
It keeps using old config and concept as much as possible.
* It only contains two major kinds of concepts:
* The "base queue": channel, levelqueue, redis
* They have the same abstraction, the same interface, and they are
tested by the same testing code.
* The "WokerPoolQueue", it uses the "base queue" to provide "worker
pool" function, calls the "handler" to process the data in the base
queue.
* The new code doesn't do "PushBack"
* Think about a queue with many workers, the "PushBack" can't guarantee
the order for re-queued unhandled items, so in new code it just does
"normal push"
* The new code doesn't do "pause/resume"
* The "pause/resume" was designed to handle some handler's failure: eg:
document indexer (elasticsearch) is down
* If a queue is paused for long time, either the producers blocks or the
new items are dropped.
* The new code doesn't do such "pause/resume" trick, it's not a common
queue's behavior and it doesn't help much.
* If there are unhandled items, the "push" function just blocks for a
few seconds and then re-queue them and retry.
* The new code doesn't do "worker booster"
* Gitea's queue's handlers are light functions, the cost is only the
go-routine, so it doesn't make sense to "boost" them.
* The new code only use "max worker number" to limit the concurrent
workers.
* The new "Push" never blocks forever
* Instead of creating more and more blocking goroutines, return an error
is more friendly to the server and to the end user.
There are more details in code comments: eg: the "Flush" problem, the
strange "code.index" hanging problem, the "immediate" queue problem.
Almost ready for review.
TODO:
* [x] add some necessary comments during review
* [x] add some more tests if necessary
* [x] update documents and config options
* [x] test max worker / active worker
* [x] re-run the CI tasks to see whether any test is flaky
* [x] improve the `handleOldLengthConfiguration` to provide more
friendly messages
* [x] fine tune default config values (eg: length?)
## Code coverage:
Change all license headers to comply with REUSE specification.
Fix#16132
Co-authored-by: flynnnnnnnnnn <flynnnnnnnnnn@github>
Co-authored-by: John Olheiser <john.olheiser@gmail.com>
Fix#19513
This PR introduce a new db method `InTransaction(context.Context)`,
and also builtin check on `db.TxContext` and `db.WithTx`.
There is also a new method `db.AutoTx` has been introduced but could be used by other PRs.
`WithTx` will always open a new transaction, if a transaction exist in context, return an error.
`AutoTx` will try to open a new transaction if no transaction exist in context.
That means it will always enter a transaction if there is no error.
Co-authored-by: delvh <dev.lh@web.de>
Co-authored-by: 6543 <6543@obermui.de>
* fix hard-coded timeout and error panic in API archive download endpoint
This commit updates the `GET /api/v1/repos/{owner}/{repo}/archive/{archive}`
endpoint which prior to this PR had a couple of issues.
1. The endpoint had a hard-coded 20s timeout for the archiver to complete after
which a 500 (Internal Server Error) was returned to client. For a scripted
API client there was no clear way of telling that the operation timed out and
that it should retry.
2. Whenever the timeout _did occur_, the code used to panic. This was caused by
the API endpoint "delegating" to the same call path as the web, which uses a
slightly different way of reporting errors (HTML rather than JSON for
example).
More specifically, `api/v1/repo/file.go#GetArchive` just called through to
`web/repo/repo.go#Download`, which expects the `Context` to have a `Render`
field set, but which is `nil` for API calls. Hence, a `nil` pointer error.
The code addresses (1) by dropping the hard-coded timeout. Instead, any
timeout/cancelation on the incoming `Context` is used.
The code addresses (2) by updating the API endpoint to use a separate call path
for the API-triggered archive download. This avoids producing HTML-errors on
errors (it now produces JSON errors).
Signed-off-by: Peter Gardfjäll <peter.gardfjall.work@gmail.com>
Within doArchive there is a service goroutine that performs the
archiving function. This goroutine reports its error using a `chan
error` called `done`. Prior to this PR this channel had 0 capacity
meaning that the goroutine would block until the `done` channel was
cleared - however there are a couple of ways in which this channel might
not be read.
The simplest solution is to add a single space of capacity to the
goroutine which will mean that the goroutine will always complete and
even if the `done` channel is not read it will be simply garbage
collected away.
(The PR also contains two other places when setting up the indexers
which do not leak but where the blocking of the sending goroutine is
also unnecessary and so we should just add a small amount of capacity
and let the sending goroutine complete as soon as it can.)
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: 6543 <6543@obermui.de>
* Start adding mechanism to return unhandled data
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Create pushback interface
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Add Pausable interface to WorkerPool and Manager
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Implement Pausable and PushBack for the bytefifos
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Implement Pausable and Pushback for ChannelQueues and ChannelUniqueQueues
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Wire in UI for pausing
Signed-off-by: Andrew Thornton <art27@cantab.net>
* add testcases and fix a few issues
Signed-off-by: Andrew Thornton <art27@cantab.net>
* fix build
Signed-off-by: Andrew Thornton <art27@cantab.net>
* prevent "race" in the test
Signed-off-by: Andrew Thornton <art27@cantab.net>
* fix jsoniter mismerge
Signed-off-by: Andrew Thornton <art27@cantab.net>
* fix conflicts
Signed-off-by: Andrew Thornton <art27@cantab.net>
* fix format
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Add warnings for no worker configurations and prevent data-loss with redis/levelqueue
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Use StopTimer
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: Lauris BH <lauris@nix.lv>
Co-authored-by: 6543 <6543@obermui.de>
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
This PR continues the work in #17125 by progressively ensuring that git
commands run within the request context.
This now means that the if there is a git repo already open in the context it will be used instead of reopening it.
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Move keys to models/keys
* Rename models/keys -> models/asymkey
* change the missed package name
* Fix package alias
* Fix test
* Fix docs
* Fix test
* Fix test
* merge
* Make modules/context.Context a context.Context
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Simplify context calls
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Set the base context for requests to the HammerContext
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
* Make archival asynchronous
The prime benefit being sought here is for large archives to not
clog up the rendering process and cause unsightly proxy timeouts.
As a secondary benefit, archive-in-progress is moved out of the
way into a /tmp file so that new archival requests for the same
commit will not get fulfilled based on an archive that isn't yet
finished.
This asynchronous system is fairly primitive; request comes in, we'll
spawn off a new goroutine to handle it, then we'll mark it as done.
Status requests will see if the file exists in the final location,
and report the archival as done when it exists.
Fixes#11265
* Archive links: drop initial delay to three-quarters of a second
Some, or perhaps even most, archives will not take all that long to archive.
The archive process starts as soon as the download button is initially
clicked, so in theory they could be done quite quickly. Drop the initial
delay down to three-quarters of a second to make it more responsive in the
common case of the archive being quickly created.
* archiver: restructure a little bit to facilitate testing
This introduces two sync.Cond pointers to the archiver package. If they're
non-nil when we go to process a request, we'll wait until signalled (at all)
to proceed. The tests will then create the sync.Cond so that it can signal
at-will and sanity-check the state of the queue at different phases.
The author believes that nil-checking these two sync.Cond pointers on every
archive processing will introduce minimal overhead with no impact on
maintainability.
* gofmt nit: no space around binary + operator
* services: archiver: appease golangci-lint, lock queueMutex
Locking/unlocking the queueMutex is allowed, but not required, for
Cond.Signal() and Cond.Broadcast(). The magic at play here is just a little
too much for golangci-lint, as we take the address of queueMutex and this is
mostly used in archiver.go; the variable still gets flagged as unused.
* archiver: tests: fix several timing nits
Once we've signaled a cond var, it may take some small amount of time for
the goroutines released to hit the spot we're wanting them to be at. Give
them an appropriate amount of time.
* archiver: tests: no underscore in var name, ungh
* archiver: tests: Test* is run in a separate context than TestMain
We must setup the mutex/cond variables at the beginning of any test that's
going to use it, or else these will be nil when the test is actually ran.
* archiver: tests: hopefully final tweak
Things got shuffled around such that we carefully build up and release
requests from the queue, so we can validate the state of the queue at each
step. Fix some assertions that no longer hold true as fallout.
* repo: Download: restore some semblance of previous behavior
When archival was made async, the GET endpoint was only useful if a previous
POST had initiated the download. This commit restores the previous behavior,
to an extent; we'll now submit the archive request there and return a
"202 Accepted" to indicate that it's processing if we didn't manage to
complete the request within ~2 seconds of submission.
This lets a client directly GET the archive, and gives them some indication
that they may attempt to GET it again at a later time.
* archiver: tests: simplify a bit further
We don't need to risk failure and use time.ParseDuration to get 2 *
time.Second.
else if isn't really necessary if the conditions are simple enough and lead
to the same result.
* archiver: tests: resolve potential source of flakiness
Increase all timeouts to 10 seconds; these aren't hard-coded sleeps, so
there's no guarantee we'll actually take that long. If we need longer to
not have a false-positive, then so be it.
While here, various assert.{Not,}Equal arguments are flipped around so that
the wording in error output reflects reality, where the expected argument is
second and actual third.
* archiver: setup infrastructure for notifying consumers of completion
This API will *not* allow consumers to subscribe to specific requests being
completed, just *any* request being completed. The caller is responsible for
determining if their request is satisfied and waiting again if needed.
* repo: archive: make GET endpoint synchronous again
If the request isn't complete, this endpoint will now submit the request and
wait for completion using the new API. This may still be susceptible to
timeouts for larger repos, but other endpoints now exist that the web
interface will use to negotiate its way through larger archive processes.
* archiver: tests: amend test to include WaitForCompletion()
This is a trivial one, so go ahead and include it.
* archiver: tests: fix test by calling NewContext()
The mutex is otherwise uninitialized, so we need to ensure that we're
actually initializing it if we plan to test it.
* archiver: tests: integrate new WaitForCompletion a little better
We can use this to wait for archives to come in, rather than spinning and
hoping with a timeout.
* archiver: tests: combine numQueued declaration with next-instruction assignment
* routers: repo: reap unused archiving flag from DownloadStatus()
This had some planned usage before, indicating whether this request
initiated the archival process or not. After several rounds of refactoring,
this use was deemed not necessary for much of anything and got boiled down
to !complete in all cases.
* services: archiver: restructure to use a channel
We now offer two forms of waiting for a request:
- WaitForCompletion: wait for completion with no timeout
- TimedWaitForCompletion: wait for completion with timeout
In both cases, we wait for the given request's cchan to close; in the latter
case, we do so with the caller-provided timeout. This completely removes the
need for busy-wait loops in Download/InitiateDownload, as it's fairly clean
to wait on a channel with timeout.
* services: archiver: use defer to unlock now that we can
This previously carried the lock into the goroutine, but an intermediate
step just added the request to archiveInProgress outside of the new
goroutine and removed the need for the goroutine to start out with it.
* Revert "archiver: tests: combine numQueued declaration with next-instruction assignment"
This reverts commit bcc5214023.
Revert "archiver: tests: integrate new WaitForCompletion a little better"
This reverts commit 9fc8bedb56.
Revert "archiver: tests: fix test by calling NewContext()"
This reverts commit 709c35685e.
Revert "archiver: tests: amend test to include WaitForCompletion()"
This reverts commit 75261f56bc.
* archiver: tests: first attempt at WaitForCompletion() tests
* archiver: tests: slight improvement, less busy-loop
Just wait for the requests to complete in order, instead of busy-waiting
with a timeout. This is slightly less fragile.
While here, reverse the arguments of a nearby assert.Equal() so that
expected/actual are correct in any test output.
* archiver: address lint nits
* services: archiver: only close the channel once
* services: archiver: use a struct{} for the wait channel
This makes it obvious that the channel is only being used as a signal,
rather than anything useful being piped through it.
* archiver: tests: fix expectations
Move the close of the channel into doArchive() itself; notably, before these
goroutines move on to waiting on the Release cond.
The tests are adjusted to reflect that we can't WaitForCompletion() after
they've already completed, as WaitForCompletion() doesn't indicate that
they've been released from the queue yet.
* archiver: tests: set cchan to nil for comparison
* archiver: move ctx.Error's back into the route handlers
We shouldn't be setting this in a service, we should just be validating the
request that we were handed.
* services: archiver: use regex to match a hash
This makes sure we don't try and use refName as a hash when it's clearly not
one, e.g. heads/pull/foo.
* routers: repo: remove the weird /archive/status endpoint
We don't need to do this anymore, we can just continue POSTing to the
archive/* endpoint until we're told the download's complete. This avoids a
potential naming conflict, where a ref could start with "status/"
* archiver: tests: bump reasonable timeout to 15s
* archiver: tests: actually release timedReq
* archiver: tests: run through inFlight instead of manually checking
While we're here, add a test for manually re-processing an archive that's
already been complete. Re-open the channel and mark it incomplete, so that
doArchive can just mark it complete again.
* initArchiveLinks: prevent default behavior from clicking
* archiver: alias gitea's context, golang context import pending
* archiver: simplify logic, just reconstruct slices
While the previous logic was perhaps slightly more efficient, the
new variant's readability is much improved.
* archiver: don't block shutdown on waiting for archive
The technique established launches a goroutine to do the wait,
which will close a wait channel upon termination. For the timeout
case, we also send back a value indicating whether the timeout was
hit or not.
The timeouts are expected to be relatively small, but still a multi-
second delay to shutdown due to this could be unfortunate.
* archiver: simplify shutdown logic
We can just grab the shutdown channel from the graceful manager instead of
constructing a channel to halt the caller and/or pass a result back.
* Style issues
* Fix mis-merge
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: Lauris BH <lauris@nix.lv>
wxiaoguang
Lunny Xiao
flynnnnnnnnnn
delvh
Peter Gardfjäll
zeripath
6543
luzpaz
Alexey 〒erentyev
John Olheiser
Kyle Evans
