Skip to main content

cmd/
standalone.rs

1// Copyright 2023 Greptime Team
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7//     http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15use std::fmt::Debug;
16use std::net::SocketAddr;
17use std::path::Path;
18use std::sync::Arc;
19use std::{fs, path};
20
21use async_trait::async_trait;
22use cache::{build_fundamental_cache_registry, with_default_composite_cache_registry};
23use catalog::information_schema::InformationExtensionRef;
24use catalog::kvbackend::{CatalogManagerConfiguratorRef, KvBackendCatalogManagerBuilder};
25use catalog::process_manager::ProcessManager;
26use clap::Parser;
27use common_base::Plugins;
28use common_catalog::consts::{MIN_USER_FLOW_ID, MIN_USER_TABLE_ID};
29use common_config::{Configurable, metadata_store_dir};
30use common_error::ext::BoxedError;
31use common_meta::cache::LayeredCacheRegistryBuilder;
32use common_meta::ddl::flow_meta::FlowMetadataAllocator;
33use common_meta::ddl::table_meta::TableMetadataAllocator;
34use common_meta::ddl::{DdlContext, NoopRegionFailureDetectorControl};
35use common_meta::ddl_manager::{DdlManager, DdlManagerConfiguratorRef, DdlManagerRef};
36use common_meta::key::flow::FlowMetadataManager;
37use common_meta::key::{TableMetadataManager, TableMetadataManagerRef};
38use common_meta::kv_backend::KvBackendRef;
39use common_meta::node_manager::{FlownodeRef, NodeManagerRef};
40use common_meta::procedure_executor::{LocalProcedureExecutor, ProcedureExecutorRef};
41use common_meta::region_keeper::MemoryRegionKeeper;
42use common_meta::region_registry::LeaderRegionRegistry;
43use common_meta::sequence::{Sequence, SequenceBuilder};
44use common_meta::wal_provider::{WalProviderRef, build_wal_provider};
45use common_options::plugin_options::StandaloneFlag;
46use common_procedure::ProcedureManagerRef;
47use common_query::prelude::set_default_prefix;
48use common_telemetry::info;
49use common_telemetry::logging::{DEFAULT_LOGGING_DIR, TracingOptions};
50use common_time::timezone::set_default_timezone;
51use common_version::{short_version, verbose_version};
52use datanode::config::DatanodeOptions;
53use datanode::datanode::{Datanode, DatanodeBuilder};
54use datanode::region_server::RegionServer;
55use flow::{
56    FlownodeBuilder, FlownodeInstance, FlownodeOptions, FrontendClient, FrontendInvoker,
57    GrpcQueryHandlerWithBoxedError,
58};
59use frontend::frontend::Frontend;
60use frontend::instance::StandaloneDatanodeManager;
61use frontend::instance::builder::FrontendBuilder;
62use frontend::server::Services;
63use meta_srv::metasrv::{FLOW_ID_SEQ, TABLE_ID_SEQ};
64use plugins::PluginOptions;
65use plugins::frontend::context::{
66    CatalogManagerConfigureContext, StandaloneCatalogManagerConfigureContext,
67};
68use plugins::standalone::context::DdlManagerConfigureContext;
69use servers::tls::{TlsMode, TlsOption, merge_tls_option};
70use snafu::ResultExt;
71use standalone::options::StandaloneOptions;
72use standalone::{StandaloneInformationExtension, StandaloneRepartitionProcedureFactory};
73use tracing_appender::non_blocking::WorkerGuard;
74
75use crate::error::{OtherSnafu, Result, StartFlownodeSnafu};
76use crate::options::{GlobalOptions, GreptimeOptions};
77use crate::{App, create_resource_limit_metrics, error, log_versions, maybe_activate_heap_profile};
78
79pub const APP_NAME: &str = "greptime-standalone";
80
81#[derive(Parser)]
82pub struct Command {
83    #[clap(subcommand)]
84    subcmd: SubCommand,
85}
86
87impl Command {
88    pub async fn build(&self, opts: GreptimeOptions<StandaloneOptions>) -> Result<Instance> {
89        self.subcmd.build(opts).await
90    }
91
92    pub fn load_options(
93        &self,
94        global_options: &GlobalOptions,
95    ) -> Result<GreptimeOptions<StandaloneOptions>> {
96        self.subcmd.load_options(global_options)
97    }
98}
99
100#[derive(Parser)]
101enum SubCommand {
102    Start(StartCommand),
103}
104
105impl SubCommand {
106    async fn build(&self, opts: GreptimeOptions<StandaloneOptions>) -> Result<Instance> {
107        match self {
108            SubCommand::Start(cmd) => cmd.build(opts).await,
109        }
110    }
111
112    fn load_options(
113        &self,
114        global_options: &GlobalOptions,
115    ) -> Result<GreptimeOptions<StandaloneOptions>> {
116        match self {
117            SubCommand::Start(cmd) => cmd.load_options(global_options),
118        }
119    }
120}
121
122pub struct Instance {
123    datanode: Datanode,
124    frontend: Frontend,
125    flownode: FlownodeInstance,
126    procedure_manager: ProcedureManagerRef,
127    wal_provider: WalProviderRef,
128    leader_services_controller: Box<dyn StandaloneLeaderServicesController>,
129    // Keep the logging guard to prevent the worker from being dropped.
130    _guard: Vec<WorkerGuard>,
131}
132
133impl Instance {
134    /// Find the socket addr of a server by its `name`.
135    pub fn server_addr(&self, name: &str) -> Option<SocketAddr> {
136        self.frontend.server_handlers().addr(name)
137    }
138
139    /// Get the mutable Frontend component of this Standalone instance for externally modification
140    /// by others (might not be in this code base, so don't delete this function).
141    pub fn mut_frontend(&mut self) -> &mut Frontend {
142        &mut self.frontend
143    }
144
145    /// Get the Datanode component of this Standalone instance for externally usage
146    /// by others (might not be in this code base, so don't delete this function).
147    pub fn datanode(&self) -> &Datanode {
148        &self.datanode
149    }
150}
151
152#[async_trait]
153impl App for Instance {
154    fn name(&self) -> &str {
155        APP_NAME
156    }
157
158    async fn start(&mut self) -> Result<()> {
159        self.datanode.start_telemetry();
160
161        self.leader_services_controller
162            .start(
163                self.procedure_manager.clone(),
164                self.wal_provider.clone(),
165                self.datanode.region_server(),
166            )
167            .await?;
168
169        plugins::start_frontend_plugins(self.frontend.instance.plugins().clone())
170            .await
171            .context(error::StartFrontendSnafu)?;
172
173        self.frontend
174            .start()
175            .await
176            .context(error::StartFrontendSnafu)?;
177
178        self.flownode.start().await.context(StartFlownodeSnafu)?;
179
180        Ok(())
181    }
182
183    async fn stop(&mut self) -> Result<()> {
184        self.frontend
185            .shutdown()
186            .await
187            .context(error::ShutdownFrontendSnafu)?;
188
189        self.leader_services_controller
190            .stop(
191                self.procedure_manager.clone(),
192                self.datanode.region_server(),
193            )
194            .await?;
195
196        self.datanode
197            .shutdown()
198            .await
199            .context(error::ShutdownDatanodeSnafu)?;
200
201        self.flownode
202            .shutdown()
203            .await
204            .context(error::ShutdownFlownodeSnafu)?;
205
206        info!("Datanode instance stopped.");
207
208        Ok(())
209    }
210}
211
212#[derive(Debug, Default, Parser)]
213pub struct StartCommand {
214    #[clap(long)]
215    http_addr: Option<String>,
216    #[clap(long = "grpc-bind-addr", alias = "rpc-bind-addr", alias = "rpc-addr")]
217    grpc_bind_addr: Option<String>,
218    #[clap(long)]
219    mysql_addr: Option<String>,
220    #[clap(long)]
221    postgres_addr: Option<String>,
222    #[clap(short, long)]
223    influxdb_enable: bool,
224    #[clap(short, long)]
225    pub config_file: Option<String>,
226    #[clap(long)]
227    tls_mode: Option<TlsMode>,
228    #[clap(long)]
229    tls_cert_path: Option<String>,
230    #[clap(long)]
231    tls_key_path: Option<String>,
232    #[clap(long)]
233    tls_watch: bool,
234    #[clap(long)]
235    user_provider: Option<String>,
236    #[clap(long, default_value = "GREPTIMEDB_STANDALONE")]
237    pub env_prefix: String,
238    /// The working home directory of this standalone instance.
239    #[clap(long)]
240    data_home: Option<String>,
241}
242
243impl StartCommand {
244    /// Load the GreptimeDB options from various sources (command line, config file or env).
245    pub fn load_options(
246        &self,
247        global_options: &GlobalOptions,
248    ) -> Result<GreptimeOptions<StandaloneOptions>> {
249        let mut opts = GreptimeOptions::<StandaloneOptions>::load_layered_options(
250            self.config_file.as_deref(),
251            self.env_prefix.as_ref(),
252        )
253        .context(error::LoadLayeredConfigSnafu)?;
254
255        self.merge_with_cli_options(global_options, &mut opts.component)?;
256        opts.component.sanitize();
257
258        Ok(opts)
259    }
260
261    // The precedence order is: cli > config file > environment variables > default values.
262    pub fn merge_with_cli_options(
263        &self,
264        global_options: &GlobalOptions,
265        opts: &mut StandaloneOptions,
266    ) -> Result<()> {
267        if let Some(dir) = &global_options.log_dir {
268            opts.logging.dir.clone_from(dir);
269        }
270
271        if global_options.log_level.is_some() {
272            opts.logging.level.clone_from(&global_options.log_level);
273        }
274
275        opts.tracing = TracingOptions {
276            #[cfg(feature = "tokio-console")]
277            tokio_console_addr: global_options.tokio_console_addr.clone(),
278        };
279
280        let tls_opts = TlsOption::new(
281            self.tls_mode,
282            self.tls_cert_path.clone(),
283            self.tls_key_path.clone(),
284            self.tls_watch,
285        );
286
287        if let Some(addr) = &self.http_addr {
288            opts.http.addr.clone_from(addr);
289        }
290
291        if let Some(data_home) = &self.data_home {
292            opts.storage.data_home.clone_from(data_home);
293        }
294
295        // If the logging dir is not set, use the default logs dir in the data home.
296        if opts.logging.dir.is_empty() {
297            opts.logging.dir = Path::new(&opts.storage.data_home)
298                .join(DEFAULT_LOGGING_DIR)
299                .to_string_lossy()
300                .to_string();
301        }
302
303        if let Some(addr) = &self.grpc_bind_addr {
304            // frontend grpc addr conflict with datanode default grpc addr
305            let datanode_grpc_addr = DatanodeOptions::default().grpc.bind_addr;
306            if addr.eq(&datanode_grpc_addr) {
307                return error::IllegalConfigSnafu {
308                    msg: format!(
309                        "gRPC listen address conflicts with datanode reserved gRPC addr: {datanode_grpc_addr}",
310                    ),
311                }.fail();
312            }
313            opts.grpc.bind_addr.clone_from(addr);
314            opts.grpc.tls = merge_tls_option(&opts.grpc.tls, tls_opts.clone());
315        }
316
317        if let Some(addr) = &self.mysql_addr {
318            opts.mysql.enable = true;
319            opts.mysql.addr.clone_from(addr);
320            opts.mysql.tls = merge_tls_option(&opts.mysql.tls, tls_opts.clone());
321        }
322
323        if let Some(addr) = &self.postgres_addr {
324            opts.postgres.enable = true;
325            opts.postgres.addr.clone_from(addr);
326            opts.postgres.tls = merge_tls_option(&opts.postgres.tls, tls_opts.clone());
327        }
328
329        if self.influxdb_enable {
330            opts.influxdb.enable = self.influxdb_enable;
331        }
332
333        if let Some(user_provider) = &self.user_provider {
334            opts.user_provider = Some(user_provider.clone());
335        }
336
337        Ok(())
338    }
339
340    #[allow(unreachable_code)]
341    #[allow(unused_variables)]
342    #[allow(clippy::diverging_sub_expression)]
343    /// Build GreptimeDB instance with the loaded options.
344    pub async fn build(&self, opts: GreptimeOptions<StandaloneOptions>) -> Result<Instance> {
345        common_runtime::init_global_runtimes(&opts.runtime);
346
347        let guard = common_telemetry::init_global_logging(
348            APP_NAME,
349            &opts.component.logging,
350            &opts.component.tracing,
351            None,
352            Some(&opts.component.slow_query),
353        );
354
355        log_versions(verbose_version(), short_version(), APP_NAME);
356        maybe_activate_heap_profile(&opts.component.memory);
357        create_resource_limit_metrics(APP_NAME);
358
359        info!("Standalone start command: {:#?}", self);
360        info!("Standalone options: {opts:#?}");
361
362        let (mut instance, _) =
363            Self::build_with(opts.component, opts.plugins, InstanceCreator::default()).await?;
364        instance._guard.extend(guard);
365        Ok(instance)
366    }
367
368    pub async fn build_with(
369        mut opts: StandaloneOptions,
370        plugin_opts: Vec<PluginOptions>,
371        creator: InstanceCreator,
372    ) -> Result<(Instance, InstanceCreatorResult)> {
373        let mut plugins = Plugins::new();
374        plugins.insert(StandaloneFlag);
375        set_default_prefix(opts.default_column_prefix.as_deref())
376            .map_err(BoxedError::new)
377            .context(error::BuildCliSnafu)?;
378
379        opts.grpc.detect_server_addr();
380        let fe_opts = opts.frontend_options();
381        let dn_opts = opts.datanode_options();
382
383        plugins::setup_frontend_plugins(&mut plugins, &plugin_opts, &fe_opts)
384            .await
385            .context(error::StartFrontendSnafu)?;
386
387        plugins::setup_datanode_plugins(&mut plugins, &plugin_opts, &dn_opts)
388            .await
389            .context(error::StartDatanodeSnafu)?;
390
391        set_default_timezone(fe_opts.default_timezone.as_deref())
392            .context(error::InitTimezoneSnafu)?;
393
394        let data_home = &dn_opts.storage.data_home;
395        // Ensure the data_home directory exists.
396        fs::create_dir_all(path::Path::new(data_home))
397            .context(error::CreateDirSnafu { dir: data_home })?;
398
399        let metadata_dir = metadata_store_dir(data_home);
400        let kv_backend = creator
401            .metadata_kv_backend_creator
402            .create(metadata_dir, &opts)
403            .await?;
404        let procedure_manager =
405            standalone::build_procedure_manager(kv_backend.clone(), opts.procedure);
406
407        plugins::setup_standalone_plugins(&mut plugins, &plugin_opts, &opts, kv_backend.clone())
408            .await
409            .context(error::SetupStandalonePluginsSnafu)?;
410
411        // Builds cache registry
412        let layered_cache_builder = LayeredCacheRegistryBuilder::default();
413        let fundamental_cache_registry = build_fundamental_cache_registry(kv_backend.clone());
414        let layered_cache_registry = Arc::new(
415            with_default_composite_cache_registry(
416                layered_cache_builder.add_cache_registry(fundamental_cache_registry),
417            )
418            .context(error::BuildCacheRegistrySnafu)?
419            .build(),
420        );
421
422        let mut builder = DatanodeBuilder::new(dn_opts, plugins.clone(), kv_backend.clone());
423        builder.with_cache_registry(layered_cache_registry.clone());
424        if let Some(writable) = creator.open_regions_writable_override {
425            builder.with_open_regions_writable_override(writable);
426        }
427        let datanode = builder.build().await.context(error::StartDatanodeSnafu)?;
428
429        let information_extension = Arc::new(StandaloneInformationExtension::new(
430            datanode.region_server(),
431            procedure_manager.clone(),
432        ));
433
434        plugins.insert::<InformationExtensionRef>(information_extension.clone());
435
436        let process_manager = Arc::new(ProcessManager::new(opts.grpc.server_addr.clone(), None));
437
438        // for standalone not use grpc, but get a handler to frontend grpc client without
439        // actually make a connection
440        let (frontend_client, frontend_instance_handler) =
441            FrontendClient::from_empty_grpc_handler(opts.query.clone());
442        let frontend_client = Arc::new(frontend_client);
443
444        let builder = KvBackendCatalogManagerBuilder::new(
445            information_extension.clone(),
446            kv_backend.clone(),
447            layered_cache_registry.clone(),
448        )
449        .with_procedure_manager(procedure_manager.clone())
450        .with_process_manager(process_manager.clone());
451        let builder = if let Some(configurator) =
452            plugins.get::<CatalogManagerConfiguratorRef<CatalogManagerConfigureContext>>()
453        {
454            let ctx = StandaloneCatalogManagerConfigureContext {
455                fe_client: frontend_client.clone(),
456            };
457            let ctx = CatalogManagerConfigureContext::Standalone(ctx);
458            configurator
459                .configure(builder, ctx)
460                .await
461                .context(OtherSnafu)?
462        } else {
463            builder
464        };
465        let catalog_manager = builder.build();
466
467        let table_metadata_manager =
468            Self::create_table_metadata_manager(kv_backend.clone()).await?;
469
470        let flow_metadata_manager = Arc::new(FlowMetadataManager::new(kv_backend.clone()));
471        let flownode_options = FlownodeOptions {
472            flow: opts.flow.clone(),
473            ..Default::default()
474        };
475
476        let flow_builder = FlownodeBuilder::new(
477            flownode_options,
478            plugins.clone(),
479            table_metadata_manager.clone(),
480            catalog_manager.clone(),
481            flow_metadata_manager.clone(),
482            frontend_client.clone(),
483        );
484        let flownode = flow_builder
485            .build()
486            .await
487            .map_err(BoxedError::new)
488            .context(error::OtherSnafu)?;
489
490        // set the ref to query for the local flow state
491        {
492            information_extension
493                .set_flow_engine(flownode.flow_engine())
494                .await;
495        }
496
497        let node_manager = creator
498            .node_manager_creator
499            .create(
500                &kv_backend,
501                datanode.region_server(),
502                flownode.flow_engine(),
503            )
504            .await?;
505
506        let table_id_allocator = creator.table_id_allocator_creator.create(&kv_backend);
507        let flow_id_sequence = Arc::new(
508            SequenceBuilder::new(FLOW_ID_SEQ, kv_backend.clone())
509                .initial(MIN_USER_FLOW_ID as u64)
510                .step(10)
511                .build(),
512        );
513        let kafka_options = opts
514            .wal
515            .clone()
516            .try_into()
517            .context(error::InvalidWalProviderSnafu)?;
518        let wal_provider = build_wal_provider(&kafka_options, kv_backend.clone())
519            .await
520            .context(error::BuildWalProviderSnafu)?;
521        let wal_provider = Arc::new(wal_provider);
522        let table_metadata_allocator = Arc::new(TableMetadataAllocator::new(
523            table_id_allocator.clone(),
524            wal_provider.clone(),
525        ));
526        let flow_metadata_allocator = Arc::new(FlowMetadataAllocator::with_noop_peer_allocator(
527            flow_id_sequence,
528        ));
529
530        let ddl_context = DdlContext {
531            node_manager: node_manager.clone(),
532            cache_invalidator: layered_cache_registry.clone(),
533            memory_region_keeper: Arc::new(MemoryRegionKeeper::default()),
534            leader_region_registry: Arc::new(LeaderRegionRegistry::default()),
535            table_metadata_manager: table_metadata_manager.clone(),
536            table_metadata_allocator: table_metadata_allocator.clone(),
537            flow_metadata_manager: flow_metadata_manager.clone(),
538            flow_metadata_allocator: flow_metadata_allocator.clone(),
539            region_failure_detector_controller: Arc::new(NoopRegionFailureDetectorControl),
540        };
541
542        let ddl_manager = DdlManager::try_new(
543            ddl_context,
544            procedure_manager.clone(),
545            Arc::new(StandaloneRepartitionProcedureFactory),
546            true,
547        )
548        .context(error::InitDdlManagerSnafu)?;
549
550        let ddl_manager = if let Some(configurator) =
551            plugins.get::<DdlManagerConfiguratorRef<DdlManagerConfigureContext>>()
552        {
553            let ctx = DdlManagerConfigureContext {
554                kv_backend: kv_backend.clone(),
555                fe_client: frontend_client.clone(),
556                catalog_manager: catalog_manager.clone(),
557            };
558            configurator
559                .configure(ddl_manager, ctx)
560                .await
561                .context(OtherSnafu)?
562        } else {
563            ddl_manager
564        };
565
566        let procedure_executor = creator
567            .procedure_executor_creator
568            .create(Arc::new(ddl_manager), procedure_manager.clone())
569            .await?;
570
571        let fe_instance = FrontendBuilder::new(
572            fe_opts.clone(),
573            kv_backend.clone(),
574            layered_cache_registry.clone(),
575            catalog_manager.clone(),
576            node_manager.clone(),
577            procedure_executor.clone(),
578            process_manager,
579        )
580        .with_plugin(plugins.clone())
581        .try_build()
582        .await
583        .context(error::StartFrontendSnafu)?;
584        let fe_instance = Arc::new(fe_instance);
585
586        // set the frontend client for flownode
587        let grpc_handler = fe_instance.clone() as Arc<dyn GrpcQueryHandlerWithBoxedError>;
588        let weak_grpc_handler = Arc::downgrade(&grpc_handler);
589        frontend_instance_handler
590            .set_handler(weak_grpc_handler)
591            .await;
592
593        // set the frontend invoker for flownode
594        let flow_streaming_engine = flownode.flow_engine().streaming_engine();
595        // flow server need to be able to use frontend to write insert requests back
596        let invoker = FrontendInvoker::build_from(
597            flow_streaming_engine.clone(),
598            catalog_manager.clone(),
599            kv_backend.clone(),
600            layered_cache_registry.clone(),
601            procedure_executor,
602            node_manager.clone(),
603        )
604        .await
605        .context(StartFlownodeSnafu)?;
606        flow_streaming_engine.set_frontend_invoker(invoker).await;
607
608        let servers = Services::new(opts, fe_instance.clone(), plugins.clone())
609            .build()
610            .context(error::StartFrontendSnafu)?;
611
612        let frontend = Frontend {
613            instance: fe_instance,
614            servers,
615            heartbeat_task: None,
616        };
617
618        let instance = Instance {
619            datanode,
620            frontend,
621            flownode,
622            procedure_manager,
623            wal_provider,
624            leader_services_controller: creator.leader_services_controller,
625            _guard: vec![],
626        };
627        let result = InstanceCreatorResult {
628            kv_backend,
629            node_manager,
630            table_id_allocator,
631        };
632        Ok((instance, result))
633    }
634
635    pub async fn create_table_metadata_manager(
636        kv_backend: KvBackendRef,
637    ) -> Result<TableMetadataManagerRef> {
638        let table_metadata_manager = Arc::new(TableMetadataManager::new(kv_backend));
639
640        table_metadata_manager
641            .init()
642            .await
643            .context(error::InitMetadataSnafu)?;
644
645        Ok(table_metadata_manager)
646    }
647}
648
649#[async_trait]
650pub trait NodeManagerCreator: Send + Sync {
651    async fn create(
652        &self,
653        kv_backend: &KvBackendRef,
654        region_server: RegionServer,
655        flow_server: FlownodeRef,
656    ) -> Result<NodeManagerRef>;
657}
658
659pub struct DefaultNodeManagerCreator;
660
661#[async_trait]
662impl NodeManagerCreator for DefaultNodeManagerCreator {
663    async fn create(
664        &self,
665        _: &KvBackendRef,
666        region_server: RegionServer,
667        flow_server: FlownodeRef,
668    ) -> Result<NodeManagerRef> {
669        Ok(Arc::new(StandaloneDatanodeManager {
670            region_server,
671            flow_server,
672        }))
673    }
674}
675
676/// Customizes how standalone opens its metadata KV backend.
677///
678/// The default implementation preserves the built-in raft-engine path. Other
679/// callers can provide a custom implementation without changing standalone
680/// configuration types.
681#[async_trait]
682pub trait MetadataKvBackendCreator: Send + Sync {
683    async fn create(&self, metadata_dir: String, opts: &StandaloneOptions) -> Result<KvBackendRef>;
684}
685
686pub struct DefaultMetadataKvBackendCreator;
687
688#[async_trait]
689impl MetadataKvBackendCreator for DefaultMetadataKvBackendCreator {
690    async fn create(&self, metadata_dir: String, opts: &StandaloneOptions) -> Result<KvBackendRef> {
691        standalone::build_metadata_kvbackend(metadata_dir, opts.metadata_store)
692            .context(error::BuildMetadataKvbackendSnafu)
693    }
694}
695
696pub trait TableIdAllocatorCreator: Send + Sync {
697    fn create(&self, kv_backend: &KvBackendRef) -> Arc<Sequence>;
698}
699
700struct DefaultTableIdAllocatorCreator;
701
702impl TableIdAllocatorCreator for DefaultTableIdAllocatorCreator {
703    fn create(&self, kv_backend: &KvBackendRef) -> Arc<Sequence> {
704        Arc::new(
705            SequenceBuilder::new(TABLE_ID_SEQ, kv_backend.clone())
706                .initial(MIN_USER_TABLE_ID as u64)
707                .step(10)
708                .build(),
709        )
710    }
711}
712
713#[async_trait]
714pub trait ProcedureExecutorCreator: Send + Sync {
715    async fn create(
716        &self,
717        ddl_manager: DdlManagerRef,
718        procedure_manager: ProcedureManagerRef,
719    ) -> Result<ProcedureExecutorRef>;
720}
721
722pub struct DefaultProcedureExecutorCreator;
723
724#[async_trait]
725impl ProcedureExecutorCreator for DefaultProcedureExecutorCreator {
726    async fn create(
727        &self,
728        ddl_manager: DdlManagerRef,
729        procedure_manager: ProcedureManagerRef,
730    ) -> Result<ProcedureExecutorRef> {
731        Ok(Arc::new(LocalProcedureExecutor::new(
732            ddl_manager,
733            procedure_manager,
734        )))
735    }
736}
737
738#[async_trait]
739pub trait StandaloneLeaderServicesController: Send + Sync {
740    /// Starts services that manage standalone metadata or WAL state.
741    ///
742    /// The default implementation starts the procedure manager and WAL provider
743    /// during instance startup.
744    async fn start(
745        &self,
746        procedure_manager: ProcedureManagerRef,
747        wal_provider: WalProviderRef,
748        region_server: RegionServer,
749    ) -> Result<()>;
750
751    /// Stops services started by [`StandaloneLeaderServicesController::start`].
752    async fn stop(
753        &self,
754        procedure_manager: ProcedureManagerRef,
755        region_server: RegionServer,
756    ) -> Result<()>;
757}
758
759pub struct DefaultStandaloneLeaderServicesController;
760
761#[async_trait]
762impl StandaloneLeaderServicesController for DefaultStandaloneLeaderServicesController {
763    async fn start(
764        &self,
765        procedure_manager: ProcedureManagerRef,
766        wal_provider: WalProviderRef,
767        _region_server: RegionServer,
768    ) -> Result<()> {
769        procedure_manager
770            .start()
771            .await
772            .context(error::StartProcedureManagerSnafu)?;
773        wal_provider
774            .start()
775            .await
776            .context(error::StartWalProviderSnafu)
777    }
778
779    async fn stop(
780        &self,
781        procedure_manager: ProcedureManagerRef,
782        _region_server: RegionServer,
783    ) -> Result<()> {
784        procedure_manager
785            .stop()
786            .await
787            .context(error::StopProcedureManagerSnafu)
788    }
789}
790
791/// `InstanceCreator` is used for grouping various component creators for building the
792/// Standalone instance, suitable for customizing how the instance can be built.
793pub struct InstanceCreator {
794    /// Hook for replacing metadata KV construction while reusing the rest of the
795    /// standalone build flow.
796    metadata_kv_backend_creator: Box<dyn MetadataKvBackendCreator>,
797    node_manager_creator: Box<dyn NodeManagerCreator>,
798    table_id_allocator_creator: Box<dyn TableIdAllocatorCreator>,
799    procedure_executor_creator: Box<dyn ProcedureExecutorCreator>,
800    leader_services_controller: Box<dyn StandaloneLeaderServicesController>,
801    open_regions_writable_override: Option<bool>,
802}
803
804impl InstanceCreator {
805    pub fn new(
806        node_manager_creator: Box<dyn NodeManagerCreator>,
807        table_id_allocator_creator: Box<dyn TableIdAllocatorCreator>,
808        procedure_executor_creator: Box<dyn ProcedureExecutorCreator>,
809    ) -> Self {
810        Self {
811            metadata_kv_backend_creator: Box::new(DefaultMetadataKvBackendCreator),
812            node_manager_creator,
813            table_id_allocator_creator,
814            procedure_executor_creator,
815            leader_services_controller: Box::new(DefaultStandaloneLeaderServicesController),
816            open_regions_writable_override: None,
817        }
818    }
819
820    pub fn with_metadata_kv_backend_creator(
821        mut self,
822        metadata_kv_backend_creator: Box<dyn MetadataKvBackendCreator>,
823    ) -> Self {
824        self.metadata_kv_backend_creator = metadata_kv_backend_creator;
825        self
826    }
827
828    /// Wraps the metadata backend creator while retaining the default creator.
829    ///
830    /// This is useful for callers that need to add runtime behavior around
831    /// metadata access without reimplementing backend selection.
832    pub fn map_metadata_kv_backend_creator<F>(mut self, f: F) -> Self
833    where
834        F: FnOnce(Box<dyn MetadataKvBackendCreator>) -> Box<dyn MetadataKvBackendCreator>,
835    {
836        self.metadata_kv_backend_creator = f(self.metadata_kv_backend_creator);
837        self
838    }
839
840    /// Wraps node-manager creation while preserving the selected standalone node manager.
841    pub fn map_node_manager_creator<F>(mut self, f: F) -> Self
842    where
843        F: FnOnce(Box<dyn NodeManagerCreator>) -> Box<dyn NodeManagerCreator>,
844    {
845        self.node_manager_creator = f(self.node_manager_creator);
846        self
847    }
848
849    /// Wraps procedure-executor creation while preserving the current setup.
850    pub fn map_procedure_executor_creator<F>(mut self, f: F) -> Self
851    where
852        F: FnOnce(Box<dyn ProcedureExecutorCreator>) -> Box<dyn ProcedureExecutorCreator>,
853    {
854        self.procedure_executor_creator = f(self.procedure_executor_creator);
855        self
856    }
857
858    /// Replaces startup/shutdown ownership for procedure manager and WAL provider.
859    pub fn with_leader_services_controller(
860        mut self,
861        leader_services_controller: Box<dyn StandaloneLeaderServicesController>,
862    ) -> Self {
863        self.leader_services_controller = leader_services_controller;
864        self
865    }
866
867    /// Overrides whether regions opened during startup should become writable.
868    ///
869    /// `None` keeps the default startup behavior (regions open writable).
870    ///
871    /// Warning: setting this to `false` in standalone mode will leave reopened regions
872    /// permanently read-only. Standalone has no metasrv heartbeat or region-role
873    /// reconciliation, so there is no path to promote regions to Leader after startup.
874    pub fn with_open_regions_writable_override(mut self, writable: bool) -> Self {
875        self.open_regions_writable_override = Some(writable);
876        self
877    }
878}
879
880impl Default for InstanceCreator {
881    fn default() -> Self {
882        Self {
883            metadata_kv_backend_creator: Box::new(DefaultMetadataKvBackendCreator),
884            node_manager_creator: Box::new(DefaultNodeManagerCreator),
885            table_id_allocator_creator: Box::new(DefaultTableIdAllocatorCreator),
886            procedure_executor_creator: Box::new(DefaultProcedureExecutorCreator),
887            leader_services_controller: Box::new(DefaultStandaloneLeaderServicesController),
888            open_regions_writable_override: None,
889        }
890    }
891}
892
893/// `InstanceCreatorResult` is expected to be used paired with [InstanceCreator].
894/// It stores the created and other important components for further reusing.
895pub struct InstanceCreatorResult {
896    pub kv_backend: KvBackendRef,
897    pub node_manager: NodeManagerRef,
898    pub table_id_allocator: Arc<Sequence>,
899}
900
901#[cfg(test)]
902mod tests {
903    use std::default::Default;
904    use std::io::Write;
905    use std::time::Duration;
906
907    use auth::{Identity, Password, UserProviderRef};
908    use clap::{CommandFactory, Parser};
909    use common_base::readable_size::ReadableSize;
910    use common_config::ENV_VAR_SEP;
911    use common_options::plugin_options::StandaloneFlag;
912    use common_test_util::temp_dir::create_named_temp_file;
913    use common_wal::config::DatanodeWalConfig;
914    use frontend::frontend::FrontendOptions;
915    use object_store::config::{FileConfig, GcsConfig};
916    use servers::grpc::GrpcOptions;
917
918    use super::*;
919    use crate::options::GlobalOptions;
920
921    #[tokio::test]
922    async fn test_try_from_start_command_to_anymap() {
923        let fe_opts = FrontendOptions {
924            user_provider: Some("static_user_provider:cmd:test=test".to_string()),
925            ..Default::default()
926        };
927
928        let mut plugins = Plugins::new();
929        plugins.insert(StandaloneFlag);
930        plugins::setup_frontend_plugins(&mut plugins, &[], &fe_opts)
931            .await
932            .unwrap();
933
934        let provider = plugins.get::<UserProviderRef>().unwrap();
935        let result = provider
936            .authenticate(
937                Identity::UserId("test", None),
938                Password::PlainText("test".to_string().into()),
939            )
940            .await;
941        let _ = result.unwrap();
942    }
943
944    #[test]
945    fn test_toml() {
946        let opts = StandaloneOptions::default();
947        let toml_string = toml::to_string(&opts).unwrap();
948        let _parsed: StandaloneOptions = toml::from_str(&toml_string).unwrap();
949    }
950
951    #[test]
952    fn test_read_from_config_file() {
953        let mut file = create_named_temp_file();
954        let toml_str = r#"
955            enable_memory_catalog = true
956
957            [wal]
958            provider = "raft_engine"
959            dir = "./greptimedb_data/test/wal"
960            file_size = "1GB"
961            purge_threshold = "50GB"
962            purge_interval = "10m"
963            read_batch_size = 128
964            sync_write = false
965
966            [storage]
967            data_home = "./greptimedb_data/"
968            type = "File"
969
970            [[storage.providers]]
971            type = "Gcs"
972            bucket = "foo"
973            endpoint = "bar"
974
975            [[storage.providers]]
976            type = "S3"
977            access_key_id = "access_key_id"
978            secret_access_key = "secret_access_key"
979
980            [storage.compaction]
981            max_inflight_tasks = 3
982            max_files_in_level0 = 7
983            max_purge_tasks = 32
984
985            [storage.manifest]
986            checkpoint_margin = 9
987            gc_duration = '7s'
988
989            [http]
990            addr = "127.0.0.1:4000"
991            timeout = "33s"
992            body_limit = "128MB"
993
994            [opentsdb]
995            enable = true
996
997            [logging]
998            level = "debug"
999            dir = "./greptimedb_data/test/logs"
1000        "#;
1001        write!(file, "{}", toml_str).unwrap();
1002        let cmd = StartCommand {
1003            config_file: Some(file.path().to_str().unwrap().to_string()),
1004            user_provider: Some("static_user_provider:cmd:test=test".to_string()),
1005            ..Default::default()
1006        };
1007
1008        let options = cmd
1009            .load_options(&GlobalOptions::default())
1010            .unwrap()
1011            .component;
1012        let fe_opts = options.frontend_options();
1013        let dn_opts = options.datanode_options();
1014        let logging_opts = options.logging;
1015        assert_eq!("127.0.0.1:4000".to_string(), fe_opts.http.addr);
1016        assert_eq!(Duration::from_secs(33), fe_opts.http.timeout);
1017        assert_eq!(ReadableSize::mb(128), fe_opts.http.body_limit);
1018        assert_eq!("127.0.0.1:4001".to_string(), fe_opts.grpc.bind_addr);
1019        assert!(fe_opts.mysql.enable);
1020        assert_eq!("127.0.0.1:4002", fe_opts.mysql.addr);
1021        assert_eq!(2, fe_opts.mysql.runtime_size);
1022        assert_eq!(None, fe_opts.mysql.reject_no_database);
1023        assert!(fe_opts.influxdb.enable);
1024        assert!(fe_opts.opentsdb.enable);
1025
1026        let DatanodeWalConfig::RaftEngine(raft_engine_config) = dn_opts.wal else {
1027            unreachable!()
1028        };
1029        assert_eq!(
1030            "./greptimedb_data/test/wal",
1031            raft_engine_config.dir.unwrap()
1032        );
1033
1034        assert!(matches!(
1035            &dn_opts.storage.store,
1036            object_store::config::ObjectStoreConfig::File(FileConfig { .. })
1037        ));
1038        assert_eq!(dn_opts.storage.providers.len(), 2);
1039        assert!(matches!(
1040            dn_opts.storage.providers[0],
1041            object_store::config::ObjectStoreConfig::Gcs(GcsConfig { .. })
1042        ));
1043        match &dn_opts.storage.providers[1] {
1044            object_store::config::ObjectStoreConfig::S3(s3_config) => {
1045                assert_eq!(
1046                    "SecretBox<alloc::string::String>([REDACTED])".to_string(),
1047                    format!("{:?}", s3_config.connection.access_key_id)
1048                );
1049            }
1050            _ => {
1051                unreachable!()
1052            }
1053        }
1054
1055        assert_eq!("debug", logging_opts.level.as_ref().unwrap());
1056        assert_eq!("./greptimedb_data/test/logs".to_string(), logging_opts.dir);
1057    }
1058
1059    #[test]
1060    fn test_load_log_options_from_cli() {
1061        let cmd = StartCommand {
1062            user_provider: Some("static_user_provider:cmd:test=test".to_string()),
1063            mysql_addr: Some("127.0.0.1:4002".to_string()),
1064            postgres_addr: Some("127.0.0.1:4003".to_string()),
1065            ..Default::default()
1066        };
1067
1068        let opts = cmd
1069            .load_options(&GlobalOptions {
1070                log_dir: Some("./greptimedb_data/test/logs".to_string()),
1071                log_level: Some("debug".to_string()),
1072
1073                #[cfg(feature = "tokio-console")]
1074                tokio_console_addr: None,
1075            })
1076            .unwrap()
1077            .component;
1078
1079        assert_eq!("./greptimedb_data/test/logs", opts.logging.dir);
1080        assert_eq!("debug", opts.logging.level.unwrap());
1081    }
1082
1083    #[test]
1084    fn test_config_precedence_order() {
1085        let mut file = create_named_temp_file();
1086        let toml_str = r#"
1087            [http]
1088            addr = "127.0.0.1:4000"
1089
1090            [logging]
1091            level = "debug"
1092        "#;
1093        write!(file, "{}", toml_str).unwrap();
1094
1095        let env_prefix = "STANDALONE_UT";
1096        temp_env::with_vars(
1097            [
1098                (
1099                    // logging.dir = /other/log/dir
1100                    [
1101                        env_prefix.to_string(),
1102                        "logging".to_uppercase(),
1103                        "dir".to_uppercase(),
1104                    ]
1105                    .join(ENV_VAR_SEP),
1106                    Some("/other/log/dir"),
1107                ),
1108                (
1109                    // logging.level = info
1110                    [
1111                        env_prefix.to_string(),
1112                        "logging".to_uppercase(),
1113                        "level".to_uppercase(),
1114                    ]
1115                    .join(ENV_VAR_SEP),
1116                    Some("info"),
1117                ),
1118                (
1119                    // http.addr = 127.0.0.1:24000
1120                    [
1121                        env_prefix.to_string(),
1122                        "http".to_uppercase(),
1123                        "addr".to_uppercase(),
1124                    ]
1125                    .join(ENV_VAR_SEP),
1126                    Some("127.0.0.1:24000"),
1127                ),
1128            ],
1129            || {
1130                let command = StartCommand {
1131                    config_file: Some(file.path().to_str().unwrap().to_string()),
1132                    http_addr: Some("127.0.0.1:14000".to_string()),
1133                    env_prefix: env_prefix.to_string(),
1134                    ..Default::default()
1135                };
1136
1137                let opts = command.load_options(&Default::default()).unwrap().component;
1138
1139                // Should be read from env, env > default values.
1140                assert_eq!(opts.logging.dir, "/other/log/dir");
1141
1142                // Should be read from config file, config file > env > default values.
1143                assert_eq!(opts.logging.level.as_ref().unwrap(), "debug");
1144
1145                // Should be read from cli, cli > config file > env > default values.
1146                let fe_opts = opts.frontend_options();
1147                assert_eq!(fe_opts.http.addr, "127.0.0.1:14000");
1148                assert_eq!(ReadableSize::mb(64), fe_opts.http.body_limit);
1149
1150                // Should be default value.
1151                assert_eq!(fe_opts.grpc.bind_addr, GrpcOptions::default().bind_addr);
1152            },
1153        );
1154    }
1155
1156    #[test]
1157    fn test_parse_grpc_bind_addr_aliases() {
1158        let command =
1159            StartCommand::try_parse_from(["standalone", "--grpc-bind-addr", "127.0.0.1:14001"])
1160                .unwrap();
1161        assert_eq!(command.grpc_bind_addr.as_deref(), Some("127.0.0.1:14001"));
1162
1163        let command =
1164            StartCommand::try_parse_from(["standalone", "--rpc-bind-addr", "127.0.0.1:24001"])
1165                .unwrap();
1166        assert_eq!(command.grpc_bind_addr.as_deref(), Some("127.0.0.1:24001"));
1167
1168        let command =
1169            StartCommand::try_parse_from(["standalone", "--rpc-addr", "127.0.0.1:34001"]).unwrap();
1170        assert_eq!(command.grpc_bind_addr.as_deref(), Some("127.0.0.1:34001"));
1171    }
1172
1173    #[test]
1174    fn test_help_uses_grpc_option_names() {
1175        let mut cmd = StartCommand::command();
1176        let mut help = Vec::new();
1177        cmd.write_long_help(&mut help).unwrap();
1178        let help = String::from_utf8(help).unwrap();
1179
1180        assert!(help.contains("--grpc-bind-addr"));
1181        assert!(!help.contains("--rpc-bind-addr"));
1182        assert!(!help.contains("--rpc-addr"));
1183    }
1184
1185    #[test]
1186    fn test_load_default_standalone_options() {
1187        let options =
1188            StandaloneOptions::load_layered_options(None, "GREPTIMEDB_STANDALONE").unwrap();
1189        let default_options = StandaloneOptions::default();
1190        assert_eq!(options.enable_telemetry, default_options.enable_telemetry);
1191        assert_eq!(options.http, default_options.http);
1192        assert_eq!(options.grpc, default_options.grpc);
1193        assert_eq!(options.mysql, default_options.mysql);
1194        assert_eq!(options.postgres, default_options.postgres);
1195        assert_eq!(options.opentsdb, default_options.opentsdb);
1196        assert_eq!(options.influxdb, default_options.influxdb);
1197        assert_eq!(options.prom_store, default_options.prom_store);
1198        assert_eq!(options.wal, default_options.wal);
1199        assert_eq!(options.metadata_store, default_options.metadata_store);
1200        assert_eq!(options.procedure, default_options.procedure);
1201        assert_eq!(options.logging, default_options.logging);
1202        assert_eq!(options.region_engine, default_options.region_engine);
1203    }
1204
1205    #[test]
1206    fn test_cache_config() {
1207        let toml_str = r#"
1208            [storage]
1209            data_home = "test_data_home"
1210            type = "S3"
1211            [storage.cache_config]
1212            enable_read_cache = true
1213        "#;
1214        let mut opts: StandaloneOptions = toml::from_str(toml_str).unwrap();
1215        opts.sanitize();
1216        assert!(opts.storage.store.cache_config().unwrap().enable_read_cache);
1217        assert_eq!(
1218            opts.storage.store.cache_config().unwrap().cache_path,
1219            "test_data_home"
1220        );
1221    }
1222}