graphql-engine/server/src-lib/Hasura/GraphQL/Transport/WSServerApp.hs

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

185 lines
6.5 KiB
Haskell
Raw Normal View History

module Hasura.GraphQL.Transport.WSServerApp
( createWSServerApp,
stopWSServerApp,
createWSServerEnv,
)
where
import Control.Concurrent.Async.Lifted.Safe qualified as LA
import Control.Concurrent.STM qualified as STM
import Control.Exception.Lifted
import Control.Monad.Trans.Control qualified as MC
import Data.Aeson (object, toJSON, (.=))
import Data.ByteString.Char8 qualified as B (pack)
import Data.Text (pack)
import Hasura.App.State
import Hasura.Backends.DataConnector.Agent.Client (AgentLicenseKey)
import Hasura.CredentialCache
import Hasura.GraphQL.Execute qualified as E
import Hasura.GraphQL.Logging
import Hasura.GraphQL.Transport.HTTP (MonadExecuteQuery)
import Hasura.GraphQL.Transport.Instances ()
import Hasura.GraphQL.Transport.WebSocket
import Hasura.GraphQL.Transport.WebSocket.Protocol
import Hasura.GraphQL.Transport.WebSocket.Server qualified as WS
import Hasura.GraphQL.Transport.WebSocket.Types
import Hasura.Logging qualified as L
import Hasura.Metadata.Class
import Hasura.Prelude
import Hasura.QueryTags
import Hasura.RQL.Types.SchemaCache
import Hasura.Server.AppStateRef
import Hasura.Server.Auth (UserAuthentication)
import Hasura.Server.Init.Config
( WSConnectionInitTimeout,
)
server: operation timeout with postgres cancelling ### Description This PR implements operation timeouts, as specced in #1232. RFC: [rfcs/operation-timeout-api-limits.md](https://github.com/hasura/graphql-engine-mono/blob/c025a90fe9779436bc0188a2bbf0ad95b5ed1f32/rfcs/operation-timeout-api-limits.md) There's still some things to be done (tests and docs most notably), but apart from that it can be reviewed. I'd still appreciate feedback on the RFC! TODO: - [x] break out the `ApiLimits` refactoring into a separate PR: #2103 - [x] finish the `pg-client-hs` PR: https://github.com/hasura/pg-client-hs/pull/39 - [x] remove configurability, after testing, prior to merging - [ ] tests: #2390 has some tests that I've run locally to confirm things work on a fundamental level - [x] changelog - [x] documentation - [x] fill in the detailed PR checklist ### Changelog - [x] `CHANGELOG.md` is updated with user-facing content relevant to this PR. If no changelog is required, then add the `no-changelog-required` label. ### Affected components - [x] Server - [ ] Console - [ ] CLI - [x] Docs - [ ] Tests ### Related Issues Product spec: #1232. ### Solution and Design Compare `rfcs/operation-timeout-api-limits.md`. ### Steps to test and verify Configure operation timeouts, e.g. by posting ``` { "type": "set_api_limits", "args": { "operation_timeout": { "global": 3 } } } ``` to `v1/metadata` to set an operation timeout of 3s. Then verify that 1. non-admin queries that take longer than 3s time out with a nice error message 2. that those queries return after ~3s (at least for postgres) 3. also that everything else still works as usual ### Limitations, known bugs & workarounds - while this will cause slow queries against any backends to fail, it's only verified to actually interrupt queries against postgres - this will only successfully short-cut (cancel) queries to postgres if the database server is responsive #### Catalog upgrade Does this PR change Hasura Catalog version? - [x] No #### Metadata Does this PR add a new Metadata feature? - [x] Yes - Does `run_sql` auto manages the new metadata through schema diffing? - [x] Not required - Does `run_sql` auto manages the definitions of metadata on renaming? - [x] Not required - Does `export_metadata`/`replace_metadata` supports the new metadata added? - [x] Yes #### GraphQL - [x] No new GraphQL schema is generated #### Breaking changes - [x] No Breaking changes PR-URL: https://github.com/hasura/graphql-engine-mono/pull/1593 GitOrigin-RevId: f0582d0be3ed9fadf89e0c4aaf96344d18331dc4
2021-09-29 19:20:06 +03:00
import Hasura.Server.Limits
import Hasura.Server.Metrics (ServerMetrics (..))
import Hasura.Server.Prometheus
( PrometheusMetrics (..),
decWebsocketConnections,
incWebsocketConnections,
)
import Hasura.Server.Types (MonadGetPolicies (..))
harmonize network manager handling ## Description ### I want to speak to the `Manager` Oh boy. This PR is both fairly straightforward and overreaching, so let's break it down. For most network access, we need a [`HTTP.Manager`](https://hackage.haskell.org/package/http-client-0.1.0.0/docs/Network-HTTP-Client-Manager.html). It is created only once, at the top level, when starting the engine, and is then threaded through the application to wherever we need to make a network call. As of main, the way we do this is not standardized: most of the GraphQL execution code passes it "manually" as a function argument throughout the code. We also have a custom monad constraint, `HasHttpManagerM`, that describes a monad's ability to provide a manager. And, finally, several parts of the code store the manager in some kind of argument structure, such as `RunT`'s `RunCtx`. This PR's first goal is to harmonize all of this: we always create the manager at the root, and we already have it when we do our very first `runReaderT`. Wouldn't it make sense for the rest of the code to not manually pass it anywhere, to not store it anywhere, but to always rely on the current monad providing it? This is, in short, what this PR does: it implements a constraint on the base monads, so that they provide the manager, and removes most explicit passing from the code. ### First come, first served One way this PR goes a tiny bit further than "just" doing the aforementioned harmonization is that it starts the process of implementing the "Services oriented architecture" roughly outlined in this [draft document](https://docs.google.com/document/d/1FAigqrST0juU1WcT4HIxJxe1iEBwTuBZodTaeUvsKqQ/edit?usp=sharing). Instead of using the existing `HasHTTPManagerM`, this PR revamps it into the `ProvidesNetwork` service. The idea is, again, that we should make all "external" dependencies of the engine, all things that the core of the engine doesn't care about, a "service". This allows us to define clear APIs for features, to choose different implementations based on which version of the engine we're running, harmonizes our many scattered monadic constraints... Which is why this service is called "Network": we can refine it, moving forward, to be the constraint that defines how all network communication is to operate, instead of relying on disparate classes constraint or hardcoded decisions. A comment in the code clarifies this intent. ### Side-effects? In my Haskell? This PR also unavoidably touches some other aspects of the codebase. One such example: it introduces `Hasura.App.AppContext`, named after `HasuraPro.Context.AppContext`: a name for the reader structure at the base level. It also transforms `Handler` from a type alias to a newtype, as `Handler` is where we actually enforce HTTP limits; but without `Handler` being a distinct type, any code path could simply do a `runExceptT $ runReader` and forget to enforce them. (As a rule of thumb, i am starting to consider any straggling `runReaderT` or `runExceptT` as a code smell: we should not stack / unstack monads haphazardly, and every layer should be an opaque `newtype` with a corresponding run function.) ## Further work In several places, i have left TODOs when i have encountered things that suggest that we should do further unrelated cleanups. I'll write down the follow-up steps, either in the aforementioned document or on slack. But, in short, at a glance, in approximate order, we could: - delete `ExecutionCtx` as it is only a subset of `ServerCtx`, and remove one more `runReaderT` call - delete `ServerConfigCtx` as it is only a subset of `ServerCtx`, and remove it from `RunCtx` - remove `ServerCtx` from `HandlerCtx`, and make it part of `AppContext`, or even make it the `AppContext` altogether (since, at least for the OSS version, `AppContext` is there again only a subset) - remove `CacheBuildParams` and `CacheBuild` altogether, as they're just a distinct stack that is a `ReaderT` on top of `IO` that contains, you guessed it, the same thing as `ServerCtx` - move `RunT` out of `RQL.Types` and rename it, since after the previous cleanups **it only contains `UserInfo`**; it could be bundled with the authentication service, made a small implementation detail in `Hasura.Server.Auth` - rename `PGMetadaStorageT` to something a bit more accurate, such as `App`, and enforce its IO base This would significantly simply our complex stack. From there, or in parallel, we can start moving existing dependencies as Services. For the purpose of supporting read replicas entitlement, we could move `MonadResolveSource` to a `SourceResolver` service, as attempted in #7653, and transform `UserAuthenticationM` into a `Authentication` service. PR-URL: https://github.com/hasura/graphql-engine-mono/pull/7736 GitOrigin-RevId: 68cce710eb9e7d752bda1ba0c49541d24df8209f
2023-02-22 18:53:52 +03:00
import Hasura.Services.Network
import Hasura.Tracing qualified as Tracing
import Network.WebSockets qualified as WS
import System.Metrics.Gauge qualified as EKG.Gauge
createWSServerApp ::
( MonadIO m,
MC.MonadBaseControl IO m,
LA.Forall (LA.Pure m),
Rewrite `Tracing` to allow for only one `TraceT` in the entire stack. This PR is on top of #7789. ### Description This PR entirely rewrites the API of the Tracing library, to make `interpTraceT` a thing of the past. Before this change, we ran traces by sticking a `TraceT` on top of whatever we were doing. This had several major drawbacks: - we were carrying a bunch of `TraceT` across the codebase, and the entire codebase had to know about it - we needed to carry a second class constraint around (`HasReporterM`) to be able to run all of those traces - we kept having to do stack rewriting with `interpTraceT`, which went from inconvenient to horrible - we had to declare several behavioral instances on `TraceT m` This PR rewrite all of `Tracing` using a more conventional model: there is ONE `TraceT` at the bottom of the stack, and there is an associated class constraint `MonadTrace`: any part of the code that happens to satisfy `MonadTrace` is able to create new traces. We NEVER have to do stack rewriting, `interpTraceT` is gone, and `TraceT` and `Reporter` become implementation details that 99% of the code is blissfully unaware of: code that needs to do tracing only needs to declare that the monad in which it operates implements `MonadTrace`. In doing so, this PR revealed **several bugs in the codebase**: places where we were expecting to trace something, but due to the default instance of `HasReporterM IO` we would actually not do anything. This PR also splits the code of `Tracing` in more byte-sized modules, with the goal of potentially moving to `server/lib` down the line. ### Remaining work This PR is a draft; what's left to do is: - [x] make Pro compile; i haven't updated `HasuraPro/Main` yet - [x] document Tracing by writing a note that explains how to use the library, and the meaning of "reporter", "trace" and "span", as well as the pitfalls - [x] discuss some of the trade-offs in the implementation, which is why i'm opening this PR already despite it not fully building yet - [x] it depends on #7789 being merged first PR-URL: https://github.com/hasura/graphql-engine-mono/pull/7791 GitOrigin-RevId: cadd32d039134c93ddbf364599a2f4dd988adea8
2023-03-13 20:37:16 +03:00
UserAuthentication m,
E.MonadGQLExecutionCheck m,
WS.MonadWSLog m,
MonadQueryLog m,
MonadExecutionLog m,
MonadExecuteQuery m,
MonadMetadataStorage m,
MonadQueryTags m,
harmonize network manager handling ## Description ### I want to speak to the `Manager` Oh boy. This PR is both fairly straightforward and overreaching, so let's break it down. For most network access, we need a [`HTTP.Manager`](https://hackage.haskell.org/package/http-client-0.1.0.0/docs/Network-HTTP-Client-Manager.html). It is created only once, at the top level, when starting the engine, and is then threaded through the application to wherever we need to make a network call. As of main, the way we do this is not standardized: most of the GraphQL execution code passes it "manually" as a function argument throughout the code. We also have a custom monad constraint, `HasHttpManagerM`, that describes a monad's ability to provide a manager. And, finally, several parts of the code store the manager in some kind of argument structure, such as `RunT`'s `RunCtx`. This PR's first goal is to harmonize all of this: we always create the manager at the root, and we already have it when we do our very first `runReaderT`. Wouldn't it make sense for the rest of the code to not manually pass it anywhere, to not store it anywhere, but to always rely on the current monad providing it? This is, in short, what this PR does: it implements a constraint on the base monads, so that they provide the manager, and removes most explicit passing from the code. ### First come, first served One way this PR goes a tiny bit further than "just" doing the aforementioned harmonization is that it starts the process of implementing the "Services oriented architecture" roughly outlined in this [draft document](https://docs.google.com/document/d/1FAigqrST0juU1WcT4HIxJxe1iEBwTuBZodTaeUvsKqQ/edit?usp=sharing). Instead of using the existing `HasHTTPManagerM`, this PR revamps it into the `ProvidesNetwork` service. The idea is, again, that we should make all "external" dependencies of the engine, all things that the core of the engine doesn't care about, a "service". This allows us to define clear APIs for features, to choose different implementations based on which version of the engine we're running, harmonizes our many scattered monadic constraints... Which is why this service is called "Network": we can refine it, moving forward, to be the constraint that defines how all network communication is to operate, instead of relying on disparate classes constraint or hardcoded decisions. A comment in the code clarifies this intent. ### Side-effects? In my Haskell? This PR also unavoidably touches some other aspects of the codebase. One such example: it introduces `Hasura.App.AppContext`, named after `HasuraPro.Context.AppContext`: a name for the reader structure at the base level. It also transforms `Handler` from a type alias to a newtype, as `Handler` is where we actually enforce HTTP limits; but without `Handler` being a distinct type, any code path could simply do a `runExceptT $ runReader` and forget to enforce them. (As a rule of thumb, i am starting to consider any straggling `runReaderT` or `runExceptT` as a code smell: we should not stack / unstack monads haphazardly, and every layer should be an opaque `newtype` with a corresponding run function.) ## Further work In several places, i have left TODOs when i have encountered things that suggest that we should do further unrelated cleanups. I'll write down the follow-up steps, either in the aforementioned document or on slack. But, in short, at a glance, in approximate order, we could: - delete `ExecutionCtx` as it is only a subset of `ServerCtx`, and remove one more `runReaderT` call - delete `ServerConfigCtx` as it is only a subset of `ServerCtx`, and remove it from `RunCtx` - remove `ServerCtx` from `HandlerCtx`, and make it part of `AppContext`, or even make it the `AppContext` altogether (since, at least for the OSS version, `AppContext` is there again only a subset) - remove `CacheBuildParams` and `CacheBuild` altogether, as they're just a distinct stack that is a `ReaderT` on top of `IO` that contains, you guessed it, the same thing as `ServerCtx` - move `RunT` out of `RQL.Types` and rename it, since after the previous cleanups **it only contains `UserInfo`**; it could be bundled with the authentication service, made a small implementation detail in `Hasura.Server.Auth` - rename `PGMetadaStorageT` to something a bit more accurate, such as `App`, and enforce its IO base This would significantly simply our complex stack. From there, or in parallel, we can start moving existing dependencies as Services. For the purpose of supporting read replicas entitlement, we could move `MonadResolveSource` to a `SourceResolver` service, as attempted in #7653, and transform `UserAuthenticationM` into a `Authentication` service. PR-URL: https://github.com/hasura/graphql-engine-mono/pull/7736 GitOrigin-RevId: 68cce710eb9e7d752bda1ba0c49541d24df8209f
2023-02-22 18:53:52 +03:00
HasResourceLimits m,
Rewrite `Tracing` to allow for only one `TraceT` in the entire stack. This PR is on top of #7789. ### Description This PR entirely rewrites the API of the Tracing library, to make `interpTraceT` a thing of the past. Before this change, we ran traces by sticking a `TraceT` on top of whatever we were doing. This had several major drawbacks: - we were carrying a bunch of `TraceT` across the codebase, and the entire codebase had to know about it - we needed to carry a second class constraint around (`HasReporterM`) to be able to run all of those traces - we kept having to do stack rewriting with `interpTraceT`, which went from inconvenient to horrible - we had to declare several behavioral instances on `TraceT m` This PR rewrite all of `Tracing` using a more conventional model: there is ONE `TraceT` at the bottom of the stack, and there is an associated class constraint `MonadTrace`: any part of the code that happens to satisfy `MonadTrace` is able to create new traces. We NEVER have to do stack rewriting, `interpTraceT` is gone, and `TraceT` and `Reporter` become implementation details that 99% of the code is blissfully unaware of: code that needs to do tracing only needs to declare that the monad in which it operates implements `MonadTrace`. In doing so, this PR revealed **several bugs in the codebase**: places where we were expecting to trace something, but due to the default instance of `HasReporterM IO` we would actually not do anything. This PR also splits the code of `Tracing` in more byte-sized modules, with the goal of potentially moving to `server/lib` down the line. ### Remaining work This PR is a draft; what's left to do is: - [x] make Pro compile; i haven't updated `HasuraPro/Main` yet - [x] document Tracing by writing a note that explains how to use the library, and the meaning of "reporter", "trace" and "span", as well as the pitfalls - [x] discuss some of the trade-offs in the implementation, which is why i'm opening this PR already despite it not fully building yet - [x] it depends on #7789 being merged first PR-URL: https://github.com/hasura/graphql-engine-mono/pull/7791 GitOrigin-RevId: cadd32d039134c93ddbf364599a2f4dd988adea8
2023-03-13 20:37:16 +03:00
ProvidesNetwork m,
Tracing.MonadTrace m,
MonadGetPolicies m
) =>
HashSet (L.EngineLogType L.Hasura) ->
WSServerEnv impl ->
WSConnectionInitTimeout ->
Maybe (CredentialCache AgentLicenseKey) ->
-- | aka generalized 'WS.ServerApp'
WS.HasuraServerApp m
createWSServerApp enabledLogTypes serverEnv connInitTimeout licenseKeyCache = \ !ipAddress !pendingConn -> do
let getMetricsConfig = scMetricsConfig <$> getSchemaCache (_wseAppStateRef serverEnv)
WS.createServerApp getMetricsConfig connInitTimeout (_wseServer serverEnv) prometheusMetrics handlers ipAddress pendingConn
where
handlers =
WS.WSHandlers
onConnHandler
onMessageHandler
onCloseHandler
logger = _wseLogger serverEnv
serverMetrics = _wseServerMetrics serverEnv
prometheusMetrics = _wsePrometheusMetrics serverEnv
getAuthMode = acAuthMode <$> getAppContext (_wseAppStateRef serverEnv)
wsActions = mkWSActions logger
-- Mask async exceptions during event processing to help maintain integrity of mutable vars:
-- here `sp` stands for sub-protocol
onConnHandler rid rh ip sp = mask_ do
liftIO $ EKG.Gauge.inc $ smWebsocketConnections serverMetrics
liftIO $ incWebsocketConnections $ pmConnections prometheusMetrics
flip runReaderT serverEnv $ onConn rid rh ip (wsActions sp)
onMessageHandler conn bs sp =
mask_
$ onMessage enabledLogTypes getAuthMode serverEnv conn bs (wsActions sp) licenseKeyCache
onCloseHandler conn = mask_ do
granularPrometheusMetricsState <- runGetPrometheusMetricsGranularity
liftIO $ EKG.Gauge.dec $ smWebsocketConnections serverMetrics
liftIO $ decWebsocketConnections $ pmConnections prometheusMetrics
onClose logger serverMetrics prometheusMetrics (_wseSubscriptionState serverEnv) conn granularPrometheusMetricsState
stopWSServerApp :: WSServerEnv impl -> IO ()
stopWSServerApp wsEnv = WS.shutdown (_wseServer wsEnv)
createWSServerEnv ::
( HasAppEnv m,
MonadIO m
) =>
AppStateRef impl ->
m (WSServerEnv impl)
createWSServerEnv appStateRef = do
AppEnv {..} <- askAppEnv
let getCorsPolicy = acCorsPolicy <$> getAppContext appStateRef
logger = _lsLogger appEnvLoggers
AppContext {acEnableAllowlist, acAuthMode, acSQLGenCtx, acExperimentalFeatures, acDefaultNamingConvention} <- liftIO $ getAppContext appStateRef
allowlist <- liftIO $ scAllowlist <$> getSchemaCache appStateRef
corsPolicy <- liftIO getCorsPolicy
wsServer <- liftIO $ STM.atomically $ WS.createWSServer acAuthMode acEnableAllowlist allowlist corsPolicy acSQLGenCtx acExperimentalFeatures acDefaultNamingConvention logger
pure
$ WSServerEnv
(_lsLogger appEnvLoggers)
appEnvSubscriptionState
appStateRef
appEnvManager
getCorsPolicy
appEnvEnableReadOnlyMode
wsServer
appEnvWebSocketKeepAlive
appEnvServerMetrics
appEnvPrometheusMetrics
appEnvTraceSamplingPolicy
mkWSActions :: L.Logger L.Hasura -> WSSubProtocol -> WS.WSActions WSConnData
mkWSActions logger subProtocol =
WS.WSActions
mkPostExecErrMessageAction
mkOnErrorMessageAction
mkConnectionCloseAction
keepAliveAction
getServerMsgType
mkAcceptRequest
fmtErrorMessage
where
mkPostExecErrMessageAction wsConn opId execErr =
sendMsg wsConn $ case subProtocol of
Apollo -> SMData $ DataMsg opId $ throwError execErr
GraphQLWS -> SMErr $ ErrorMsg opId $ toJSON execErr
mkOnErrorMessageAction wsConn err mErrMsg =
case subProtocol of
Apollo ->
case mErrMsg of
WS.ConnInitFailed -> sendCloseWithMsg logger wsConn (WS.mkWSServerErrorCode mErrMsg err) (Just $ SMConnErr err) Nothing
WS.ClientMessageParseFailed -> sendMsg wsConn $ SMConnErr err
GraphQLWS -> sendCloseWithMsg logger wsConn (WS.mkWSServerErrorCode mErrMsg err) (Just $ SMConnErr err) Nothing
mkConnectionCloseAction wsConn opId errMsg =
when (subProtocol == GraphQLWS)
$ sendCloseWithMsg logger wsConn (GenericError4400 errMsg) (Just . SMErr $ ErrorMsg opId $ toJSON (pack errMsg)) (Just 1000)
getServerMsgType = case subProtocol of
Apollo -> SMData
GraphQLWS -> SMNext
keepAliveAction wsConn = sendMsg wsConn
$ case subProtocol of
Apollo -> SMConnKeepAlive
GraphQLWS -> SMPing . Just $ keepAliveMessage
mkAcceptRequest =
WS.defaultAcceptRequest
{ WS.acceptSubprotocol = Just . B.pack . showSubProtocol $ subProtocol
}
fmtErrorMessage errMsgs = case subProtocol of
Apollo -> object ["errors" .= errMsgs]
GraphQLWS -> toJSON errMsgs