There is a fundamental design problem of the "manager" and the "wait
group".
If nothing has started, the "Wait" just panics: sync: WaitGroup is
reused before previous Wait has returned
There is no clear solution besides a complete rewriting of the "manager"
If there are some mistakes in the app.ini, end users would just see the
"panic", but not the real error messages. A real case: #27643
This PR is just a quick fix for the annoying panic problem.
This is a simple PR which moves the `GetListener` function to a
`DefaultGetListener` function, and changes `GetListener` to be a
variable which by default points to the `DefaultGetListener` function.
This allows people who may exist quasi-downstream of Gitea to create
alternate "GetListener" functions, with identical signatures, which
return different implementations of the `net.Listener` interface. This
approach is expressly intended to be non-invasive and have the least
possible impact on the gitea codebase. A previous version of this idea
was rejected before: https://github.com/go-gitea/gitea/issues/15544 but
because of issues like: https://github.com/go-gitea/gitea/issues/22335 I
**really** think that recommending people configure proxies by hand is
exactly the wrong way to do things(This is why there is a Tor Browser.).
This tiny change lets me put proper hidden service configuration into
single `i2p.go` file which lives in `modules/graceful/` and which never
has to be checked in to your codebase or affect your dependencies or
bloat your project in any way, it can live on a branch in my fork and
I'll fast-forward every release and never the twain shall meet.
The main use-case for this is to listen on Peer-to-Peer networks and
Hidden Services directly without error-prone and cumbersome
port-forwarding configuration. For instance, I might implement an
"I2PGetListener" as follows:
```Go
// adapted from i2p.go which is unchecked-in in my modules/graceful/ directory
import "github.com/eyedeekay/onramp"
var garlic = &onramp.Garlic{}
func I2PGetListener(network, address string) (net.Listener, error) {
// Add a deferral to say that we've tried to grab a listener
defer GetManager().InformCleanup()
switch network {
case "tcp", "tcp4", "tcp6", "i2p", "i2pt":
return garlic.Listen()
case "unix", "unixpacket":
// I2P isn't really a replacement for the stuff you use Unix sockets for and it's also not an anonymity risk, so treat them normally
unixAddr, err := net.ResolveUnixAddr(network, address)
if err != nil {
return nil, err
}
return GetListenerUnix(network, unixAddr)
default:
return nil, net.UnknownNetworkError(network)
}
}
```
I could then substitute that GetListener function and be 50% of the way
to having a fully-functioning gitea-over-hidden-services instance
without any additional configuration(The other 50% doesn't require any
code-changes on gitea's part).
There are 2 advantages here, one being convenience, first this turns
hidden services into a zero-configuration option for self-hosting gitea,
and second safety, these Go libraries are passing around
hidden-service-only versions of the net.Addr struct, they're using
hidden-service-only versions of the sockets, which are both expressly
designed to never require access to any information outside the hidden
service network, manipulating the application so it reveals information
about the host becomes much more difficult, and some attacks become
nearly impossible. It also opens up TLS-over-Hidden Services support
which is niche right now, of course, but in a future where gitea
instances federate if hidden services want to be part of the federation
they're probably going to need TLS certificates. They don't need to be
painful to set up.
This doesn't fix an open issue, but it might affect:
- https://github.com/go-gitea/gitea/issues/22335 - my `i2p.go` file
actually has a mod that fixes this but it requires adding a handful of
new dependencies to gitea and isn't compatible with the normal way you
guys recommend using a proxy so I don't think it's ready to send to you
as a PR, but if I can find a non-invasive way to fix it I will.
- https://github.com/go-gitea/gitea/issues/18240
I hereby agree to the Code of Conduct published here:
8b89563bf1/CODE_OF_CONDUCT.md
I have read and understood the recommendations published here:
8b89563bf1/CONTRIBUTING.md
Thank you for your consideration.
---------
Co-authored-by: eyedeekay <idk@mulder>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Before there was a "graceful function": RunWithShutdownFns, it's mainly
for some modules which doesn't support context.
The old queue system doesn't work well with context, so the old queues
need it.
After the queue refactoring, the new queue works with context well, so,
use Golang context as much as possible, the `RunWithShutdownFns` could
be removed (replaced by RunWithCancel for context cancel mechanism), the
related code could be simplified.
This PR also fixes some legacy queue-init problems, eg:
* typo : archiver: "unable to create codes indexer queue" => "unable to
create repo-archive queue"
* no nil check for failed queues, which causes unfriendly panic
After this PR, many goroutines could have better display name:
## ⚠️ Breaking
The `log.<mode>.<logger>` style config has been dropped. If you used it,
please check the new config manual & app.example.ini to make your
instance output logs as expected.
Although many legacy options still work, it's encouraged to upgrade to
the new options.
The SMTP logger is deleted because SMTP is not suitable to collect logs.
If you have manually configured Gitea log options, please confirm the
logger system works as expected after upgrading.
## Description
Close#12082 and maybe more log-related issues, resolve some related
FIXMEs in old code (which seems unfixable before)
Just like rewriting queue #24505 : make code maintainable, clear legacy
bugs, and add the ability to support more writers (eg: JSON, structured
log)
There is a new document (with examples): `logging-config.en-us.md`
This PR is safer than the queue rewriting, because it's just for
logging, it won't break other logic.
## The old problems
The logging system is quite old and difficult to maintain:
* Unclear concepts: Logger, NamedLogger, MultiChannelledLogger,
SubLogger, EventLogger, WriterLogger etc
* Some code is diffuclt to konw whether it is right:
`log.DelNamedLogger("console")` vs `log.DelNamedLogger(log.DEFAULT)` vs
`log.DelLogger("console")`
* The old system heavily depends on ini config system, it's difficult to
create new logger for different purpose, and it's very fragile.
* The "color" trick is difficult to use and read, many colors are
unnecessary, and in the future structured log could help
* It's difficult to add other log formats, eg: JSON format
* The log outputer doesn't have full control of its goroutine, it's
difficult to make outputer have advanced behaviors
* The logs could be lost in some cases: eg: no Fatal error when using
CLI.
* Config options are passed by JSON, which is quite fragile.
* INI package makes the KEY in `[log]` section visible in `[log.sub1]`
and `[log.sub1.subA]`, this behavior is quite fragile and would cause
more unclear problems, and there is no strong requirement to support
`log.<mode>.<logger>` syntax.
## The new design
See `logger.go` for documents.
## Screenshot
<details>
</details>
## TODO
* [x] add some new tests
* [x] fix some tests
* [x] test some sub-commands (manually ....)
---------
Co-authored-by: Jason Song <i@wolfogre.com>
Co-authored-by: delvh <dev.lh@web.de>
Co-authored-by: Giteabot <teabot@gitea.io>
This PR adds support for the systemd notify protocol. Several status
messagess are provided. We should likely add a common notify/status
message for graceful.
Replaces #21140
Signed-off-by: Andrew Thornton <art27@cantab.net>
---------
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: ltdk <usr@ltdk.xyz>
Co-authored-by: Giteabot <teabot@gitea.io>
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>
This PR adds functionality to allow Gitea to sit behind an
HAProxy and HAProxy protocolled connections directly.
Fix#7508
Signed-off-by: Andrew Thornton <art27@cantab.net>
The graceful manager waits for 4 listeners to be created or to be told that they are not needed. If it is not told about them it will indefinitely and timeout.
This leads to SVC hosts not being told of being in the readyState but on Unix would lead to the termination of the process.
There was an unfortunate regression in #20299 which missed this subtly and in the case whereby SSH is disabled the `builtinUnused()` is not called.
This PR adds a call to `builtinUnused()` when not using the builtin ssh to allow `createServerWaitGroup.Done()` to be called.
In addition it was noted that the if/else clauses for timeout informing of the SVC host were in the wrong order. These have been swapped.
Fix#20609
By default Gitea will always unlink any sockets that are provided using the `LISTEN_FDS` environment variable. This is because it uses this variable to handle passing when it is doing a graceful restart. However, this same mechanism is used by systemd - which explicitly expects that passed in sockets should not be unlinked by the receiving process.
This PR adjusts Gitea's graceful restart mechanism to use an additional environment variable which tracks if a listening socket was opened by Gitea - and therefore should be unlinked on shutdown by Gitea.
Fix#20490
Co-authored-by: zeripath <art27@cantab.net>
Continues on from #19202.
Following the addition of pprof labels we can now more easily understand the relationship between a goroutine and the requests that spawn them.
This PR takes advantage of the labels and adds a few others, then provides a mechanism for the monitoring page to query the pprof goroutine profile.
The binary profile that results from this profile is immediately piped in to the google library for parsing this and then stack traces are formed for the goroutines.
If the goroutine is within a context or has been created from a goroutine within a process context it will acquire the process description labels for that process.
The goroutines are mapped with there associate pids and any that do not have an associated pid are placed in a group at the bottom as unbound.
In this way we should be able to more easily examine goroutines that have been stuck.
A manager command `gitea manager processes` is also provided that can export the processes (with or without stacktraces) to the command line.
Signed-off-by: Andrew Thornton <art27@cantab.net>
Currently Gitea will wait for HammerTime or nice shutdown if kill -1 or kill -2
is sent. We should just immediately hammer if there is a second kill.
Signed-off-by: Andrew Thornton <art27@cantab.net>
The io/ioutil package has been deprecated as of Go 1.16, see
https://golang.org/doc/go1.16#ioutil. This commit replaces the existing
io/ioutil functions with their new definitions in io and os packages.
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
* Set self-adjusting deadline for connection writing
In #16055 it appears that the simple 5s deadline doesn't work for large
file writes. Now we can't - or at least shouldn't just set no deadline
as go will happily let these connections block indefinitely. However,
what seems reasonable is to set some minimum rate we expect for writing.
This PR suggests the following algorithm:
* Every write has a minimum timeout of 5s (adjustable at compile time.)
* If there has been a previous write - then consider its previous
deadline, add half of the minimum timeout + 2s per kb about to written.
* If that new deadline is after the minimum timeout use that.
Fix#16055
* Linearly increase timeout
* Make PerWriteTimeout, PerWritePerKbTimeouts configurable
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: 6543 <6543@obermui.de>
* 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>
* move shutdownfns, terminatefns and hammerfns out of separate goroutines
Coalesce the shutdownfns etc into a list of functions that get run at shutdown
rather then have them run at goroutines blocked on selects.
This may help reduce the background select/poll load in certain
configurations.
* The LevelDB queues can actually wait on empty instead of polling
Slight refactor to cause leveldb queues to wait on empty instead of polling.
* Shutdown the shadow level queue once it is empty
* Remove bytefifo additional goroutine for readToChan as it can just be run in run
* Remove additional removeWorkers goroutine for workers
* Simplify the AtShutdown and AtTerminate functions and add Channel Flusher
* Add shutdown flusher to CUQ
* move persistable channel shutdown stuff to Shutdown Fn
* Ensure that UPCQ has the correct config
* handle shutdown during the flushing
* reduce risk of race between zeroBoost and addWorkers
* prevent double shutdown
Signed-off-by: Andrew Thornton <art27@cantab.net>
In #15826 it has become apparent that there are a few occasions when a response can
hang during writing, and because there is no timeout go will happily just block
interminably. This PR adds a fixed 5 second timeout to all writes to a connection.
Fix#15826
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Drop back to use IsAnInteractiveSession for SVC
There is an apparent permission change problem when using
IsWindowsService to determine if the SVC manager should be
used.
This PR simply drops back to using IsAnInteractiveSession as
this does not change behaviour.
Fix#15454
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Yes staticcheck I know this is deprecated
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Just leave me alone lint
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: 6543 <6543@obermui.de>
Unfortunately #14623 changed from the deprecated IsInteractiveSession to
IsWindowsService without recognising that they are the complement of
each other.
This means that Windows SVC control is not working correctly. This PR
adds some Tracing statements but also fixes the bug.
Fix#15159
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
* Make SSH server host key path configurable
* make it possible to have multiple keys
* Make gitea.rsa the default key
* Add some more logging
Signed-off-by: Andrew Thornton <art27@cantab.net>
* fix build
* take flash error message back and fix more windows lint error
* performance optimization
* own step to check lint for windows
Co-authored-by: 6543 <6543@obermui.de>
Replace all calls to os.Remove/os.RemoveAll by retrying util.Remove/util.RemoveAll and remove circular dependencies from util.
Fix#12339
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: silverwind <me@silverwind.io>
* Make LogDescriptions race safe
* Add manager commands for pausing, resuming, adding and removing loggers
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Placate lint
* Ensure that file logger is run!
* Add support for smtp and conn
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Add release-and-reopen
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
Co-authored-by: Lauris BH <lauris@nix.lv>
* Handle panics in graceful goroutines
Adds a some deferred functions to handle panics in graceful goroutines
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Handle panic in webhook.Deliver
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Handle panic in mirror.syncMirror
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: Lauris BH <lauris@nix.lv>
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
* Prevent panic during wrappedConn close at hammertime
Signed-off-by: Andrew Thornton <art27@cantab.net>
* Update modules/graceful/server.go
* Fix extraneous debug in goldmark.go
Signed-off-by: Andrew Thornton <art27@cantab.net>
silverwind