Fix shutdown hanging if the db went away. gives shutdowns a time out. improves db shutdown logging (#81813)
terry right now:
```
[20:11:31] Runtime in code/controllers/subsystem/dbcore.dm,353: Attempting to create a new db query during the world shutdown
...
[21:25:29] Runtime in code/controllers/subsystem/dbcore.dm,353: Attempting to create a new db query during the world shutdown
```
Co-authored-by: Kyle Spier-Swenson <kyleshome@gmail.com>
* refactor db connection timeouts (#81816)
Give it exponential back off and smerter logic.
* refactor db connection timeouts
---------
Co-authored-by: Kyle Spier-Swenson <kyleshome@gmail.com>
* Wait for db queries to finish during db shutdown (#77311)
* Try waiting for db queries to finish during db shutdown
---------
Co-authored-by: oranges <email@oranges.net.nz>
* Stop stack tracing if ezdb daemon is already running (#76555)
## About The Pull Request
This is an annoyance for me because I often kill the game but not the
EZDB daemon (leaving the cmd window open), and it just triggers this
stack trace while I have a debugger running. It's still able to connect
to the daemon in-game and functions perfectly fine, just hate dealing
with a completely needless stack_trace() every single time.
## Why It's Good For The Game
LET ME WORK!!! PLEASE!!!
## Changelog
Nothing that players should care about.
* Stop stack tracing if ezdb daemon is already running
---------
Co-authored-by: san7890 <the@san7890.com>
* RUNLEVEL_INIT does not exist (#75023)
This define shouldn't have been added (by me when I created runlevels)
people get it confused with RUNLEVEL_LOBBY and misuse it, like in this
verb subsystem I have no knowledge of.
- Removes RUNLEVEL_INIT.
- Fixed SSverb_manager not running during the lobby phase.
- Fixed Master.SetRunLevel having the potential to accept a scuffed
runlevel.
- Other standardizing cleanups to runlevels usage.
* RUNLEVEL_INIT does not exist
---------
Co-authored-by: Jordan Dominion <Cyberboss@users.noreply.github.com>
* Refactors and defuckulates dbcore. Adds support for min_threads rustg setting, Reduce query delay, Make unit tests faster (#74852)
dbcore was very fuckulated.
It had 3 lists of queries, but they all had their own current_run style
list to support mc_tick_check (as it was already being done before with
the undeleted query check, so i can understand why they ~~cargo culted~~
mirrored the behavior) This was silly and confusing and unneeded given
two of those loops can only process at most 25 items at a time on
default config, plus these were cheap operations (ask rustg to start
thread, ask rustg to check on thread).
Because of the confusingness of the 6 lists for 3 query states, The code
to run pending/queued queries immediately during world shutdown was
instead looking at the current_run list for active queries, **meaning
those queries got ran twice.**
The queued query system only checked the current active query count in
fire(), meaning even when there was nothing going on in this subsystem
new queries had to wait for the next fire() to run (10 ticks, so 500ms
on default config)
Those have all been fixed.
the config `BSQL_THREAD_LIMIT` has been renamed to
`POOLING_MAX_SQL_CONNECTIONS` and its default was lowered to match
MAX_CONCURRENT_QUERIES .
added a new config `POOLING_MIN_SQL_CONNECTIONS`, allowing you to
pre-allocate a reserve of sql threads.
The queue processing part of SSdbcore's fire() has been made to not obey
mc_tick_check for clarity and to make the following change easier to do:
If there is less than `MAX_CONCURRENT_QUERIES` in the active queue, new
queries activate immediately.
(its ok that there are two configs that kinda do the same thing,
POOLING_MAX_SQL_CONNECTIONS maps to max-threads in the mysql crate, and
it seems to only be a suggestion, meanwhile MAX_CONCURRENT_QUERIES can't
do anything during init, which is when the highest amount of concurrent
queries tend to happen.)
🆑
config: database configs have been updated for better control over the
connection pool
server: BSQL_THREAD_LIMIT has been renamed to
POOLING_MAX_SQL_CONNECTIONS, old configs will whine but still work.
fix: fixed rare race condition that could lead to a sql query being ran
twice during world shutdown.
/🆑
I have not tested this pr.
* Refactors and defuckulates dbcore. Adds support for min_threads rustg setting, Reduce query delay, Make unit tests faster
---------
Co-authored-by: Kyle Spier-Swenson <kyleshome@gmail.com>
* Change admin message for undeleted queries to stack trace. (#70589)
Change admin message to stack trace
The admins don't know what to do with this error and I swear to god if I get another ping without a round id or server name I am going to literally explode
* Change admin message for undeleted queries to stack trace.
Co-authored-by: oranges <email@oranges.net.nz>
* Minor QuerySelect code improvements, QuerySelect documented. (#69383)
* Minor QuerySelect improvements coded from the porcelain throne.
We don't handle bad values being given in the query list well enough. This normally won't matter, runtimes are runtimes, but if mixed with real queries, it can lead to inconsistent state where we have query datums that have been ran, and query datums that have not been ran.
pre-checking is also an option, so that it can just refuse to run any of them if one is bad by checking before, but my main goal was to prevent runtimes from bad inputs leading to undeleted queries spam while being more clear about where the bug is, not try to perfectly handle the side effects of bad code.
(To be clear, i started intending to just add codedoc but now it uses as anything and typechecks because that was just eating at me while typing up the codedoc.)
* Update code/controllers/subsystem/dbcore.dm
Co-authored-by: Mothblocks <35135081+Mothblocks@ users.noreply.github.com>
* Minor QuerySelect code improvements, QuerySelect documented.
Co-authored-by: Kyle Spier-Swenson <kyleshome@gmail.com>
Co-authored-by: Mothblocks <35135081+Mothblocks@ users.noreply.github.com>
* Adds a new config entry for SQL game log bundle query size
Also adds documentation to existing config entries that don't currently appear in any default entry files on the repo.
* Adds a new config file for database-related configs for Skyrat
* Adds a `add_log_to_mass_insert_queue` proc to handle bundling insert queries into bigger, less frequent queries
* Modifies log_message() to utilise the new add_log_to_mass_insert_queue() proc
* Actually makes the MassInsert() call asynchronous, to completely remove the freezes
* Adds code to handle sending all unsent logs to the database before the subsystem is shutdown, to reduce the loss of logs at the end of a round
* Adds MC initialization stages. Earlier stages can fire while later ones init. Fixes tgui chat reconnection banner showing during init. (#66473)
* Adds MC initialization stages. Earlier stages can fire while later ones init.
Removes TICK_LIMIT_MC_INIT config for barely doing anything to speed up init and being inconvenient to work with if fires and inits can happen at the same time.
* Adds MC initialization stages. Earlier stages can fire while later ones init. Fixes tgui chat reconnection banner showing during init.
Co-authored-by: Kyle Spier-Swenson <kyleshome@gmail.com>
* Refactors dbcore and limits the maximum amount of concurrent async queries to a variable amount (#59676)
Refactors dbcore to work off a subsystem if executed async and limits the maximum amount of concurrent async queries to 25.
This has been tested locally on a mysql docker image and there were no crashes (as long as you didn't run it with debug extools) + data was getting recorded fine.
Why It's Good For The Game
May or may not resolve terry crashes, however, each query creates a new thread which takes up 2mb, preventing the game from using that 2mb. This can lead to ooms if they stack up, e.g. due to poor connectivity. This solves that issue.
maintainer note: this did not actually resolve the crashes, but has value anyway. Crashes were sidestepped fixed by finding out Large Address Awareness works
cl
refactor: Refactors dbcore.dm to possibly resolve the crashes that happen on Terry.
/cl
* Fixes an oversight in database code and cleans up telemetry (#64177)
As it is right now, we never actually clear the temporary list processing_queries
So if the subsystem is for some reason unable to complete a run, we will just whip right back around to it again
If it's been long enough, this could even cause horrific log spam. There was just now a manuel round with roughly 30k undeleted query errors. not good.
But what was actually not deleting you may ask?
Well
When you create a db request, a 5 minute timer starts. after those 5 minutes are up, the request is qdeleted by the db subsystem
This is to prevent the creation of unused requests, and to handle requests that are never cleaned up
Telemetry code was creating all of its db requests inside a for loop that could check tick, and then later
attempting to call them in series
Since requests by default sleep, this almost always lead to undeleted queries, which harddel'd given long enough periods
I've fixed this by moving the data gathering away from the query creation
Why is it good for the game
I was working on atmos code, happy, safe in my delusion, when suddenly I got a ping from tattle freaking out over 200 undeleted queries a second
This resolves that issue, so I can once again live in peace
Changelog
cl
admin: Telemetry code will spam you with undeleted query logs much less often now!
server: Improved how the db subsystem handles undeleted queries, should never have an incident like that again
/cl
* Fixes an error in telemetry queries (#64205)
* Hardsynced time_track.dm with upstream
Co-authored-by: Watermelon914 <37270891+Watermelon914@users.noreply.github.com>
Co-authored-by: LemonInTheDark <58055496+LemonInTheDark@users.noreply.github.com>
* Revert "[MIRROR] Refactors dbcore and limits the maximum amount of concurrent async queries to a variable amount (#6436)"
This reverts commit e2519525f9.
* Update time_track.dm
Co-authored-by: Gandalf <jzo123@hotmail.com>
* Refactors dbcore and limits the maximum amount of concurrent async queries to a variable amount (#59676)
Refactors dbcore to work off a subsystem if executed async and limits the maximum amount of concurrent async queries to 25.
This has been tested locally on a mysql docker image and there were no crashes (as long as you didn't run it with debug extools) + data was getting recorded fine.
Why It's Good For The Game
May or may not resolve terry crashes, however, each query creates a new thread which takes up 2mb, preventing the game from using that 2mb. This can lead to ooms if they stack up, e.g. due to poor connectivity. This solves that issue.
maintainer note: this did not actually resolve the crashes, but has value anyway. Crashes were sidestepped fixed by finding out Large Address Awareness works
cl
refactor: Refactors dbcore.dm to possibly resolve the crashes that happen on Terry.
/cl
* Refactors dbcore and limits the maximum amount of concurrent async queries to a variable amount
Co-authored-by: Watermelon914 <37270891+Watermelon914@users.noreply.github.com>
* Better accomodation of multiple servers using the same database.
* updated the PR number in readme.md
* Updated the database_changelog description.
* Oopsie.
* Cosmetic touch and sql query fixes.
* Opsie wopsie
* Oopsie x2
* Fugg
* Case of lower
* More changes
* Ruins the nice 420 diff, brainfart when doing the second batch of conversions
* More changes
* Next batch. I think
* Converts even more paths
* Restarts bots
* Capital Free Zone
* Come on travis, do something
* Renames areas
* Bots, please stop dying
* Updates CONTRIBUTING.md and updates a few paths I missed.
* APC recgarftzfvas
/obj/item/computer_hardware/recharger/apc to /obj/item/computer_hardware/recharger/apc_recharger
cl
refactor: BSQL has been replaced with rust-g, improving SQL reliability and safety.
/cl
Queries were parameterized as much as possible. Careful code read by someone else also versed in SQL would be a good idea. Things seemed fine spot-testing roundstart, library, bans, and polls, but more stress-testing may be useful.
Also updates rust-g to v0.4.5:
Improved Linux .so detection
Perlin noise functions
Also updates Travis script to bionic.
cl Naksu
code: DB queries are now async by default, the code that runs in /world/New has been adjusted to continue blocking as it apparently is not allowed to sleep.
/cl
closes#40610
* Add BSQL library v1.1.1.0
* Modify dbcore to use BSQL
* Add missing QDEL_NULL for connectOperation
* Moves BSQL_Shutdown() call to dbcore shutdown
* Fix passing the wrong argument to DBQuery/New()
* Darn it @Jordie0608. Fixes db calls without Connect check
No seriously please make sure I'm not breaking anything
* Queries with a null connection won't runtime
* Fix quoting
* Fix mistake
* Update BSQL to v1.1.2.0
* Update BSQL DMAPI to v1.0.1.0
* Fix connection instatiation
* Does the smart thing in regards to quoting
* Fix braces
* Update BSQL to 1.2.0.0. DMAPI to 1.1.0.0
* Execute/NextRow/MassInsert now have async parameter
* Build BSQL for tests
* Add missing apt source
* Def still need gcc-multilib
* Wut
* Revert "Wut"
This reverts commit d7c98a9a6b27f6db03e9f5cc534650d59d018048.
* Try this then
* Could it really be that simple?
* Literally running out of ideas here
* Update BSQL to v1.2.1.0 DMAPI to v1.1.1.0
* Update BSQL travis version
* Nothing about this makes sense tbqhwyfam
* Whoo boy
* No idea why this isn't working tbh
* Absolute madness
* Ahhhhhhhhhhhhh
* *deep breath*
* "though yet again i was frustrated by failure"
* Add BSQL to Dockerfile
* Pass through MassInsert async param
* BSQL to v1.3.0.0 DMAPI to 1.2.0.0
* Add timeout support
* Wait, something's fucky
* Wtf is this meme?
* Just get good lmao
* Just stop being shit lol
* Stupid verbose logging
* Remove verbosity
* Good god
* BSQL to v1.3.0.1 DMAPI to v1.2.0.1
* BSQL to v1.3.0.2
* Update BSQL travis version
* Update BSQL docker version
* Didn't mean to change that
* Strip connection information from debug logs and make it configgable
* Move this to where CONFIG_GET is defined
* Qdels all queries, adds sleep handling
* DB Core messages admins about undeleted queries
* Compile fixes. Adds missing set waitfor
* Remove world/New shennanigans. Add DBQuery/BlockingExecute()
* Less spammy notifications to admins about undeleted queries
* Increase dbcore fire time to 1 minute
* Upgrade undeleted query warning
* Better place of death
* Fix build
* Remove BlockingExecute, see BSQL PR for why
* Yep, missed that one.
* Psyche, that's the WRONG QUERY!!
For planned change to how TGS initializes server instances ahead of time so rounds aren't tracked as being far longer than they really would be. Similarly end time split for a more accurate round duration.
