com.transferwise.tasks.TasksProperties Maven / Gradle / Ivy
Show all versions of tw-tasks-core Show documentation
package com.transferwise.tasks;
import com.transferwise.tasks.config.ResolvedValue;
import com.transferwise.tasks.utils.ClientIdUtils;
import java.time.Duration;
import java.util.ArrayList;
import java.util.Collections;
import java.util.HashMap;
import java.util.HashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;
import javax.validation.Valid;
import javax.validation.constraints.Min;
import javax.validation.constraints.NotBlank;
import javax.validation.constraints.NotEmpty;
import javax.validation.constraints.NotNull;
import javax.validation.constraints.Positive;
import lombok.Data;
import lombok.experimental.Accessors;
@Data
public class TasksProperties {
/**
* Unique id for service in the whole Company infrastructure.
*/
@NotBlank
@ResolvedValue
private String groupId;
/**
* Unique node id in the service cluster. It helps to make crash recovery for a node very fast, but also is good for logging and tracking reasons.
*/
@NotBlank
@ResolvedValue
private String clientId = ClientIdUtils.clientIdFromHostname();
/**
* How often do we check if any task is stuck.
*/
@NotNull
private Duration stuckTasksPollingInterval = Duration.ofMinutes(1);
/**
* How often do we try to clean very old tasks from the database.
*/
@NotNull
private Duration tasksCleaningInterval = Duration.ofSeconds(1);
/**
* How often do we check if any scheduled task should be executed now.
*/
@NotNull
private Duration waitingTasksPollingInterval = Duration.ofSeconds(5);
/**
* Generic maximum time to wait for any lock, event or polling. It helps to make the system more robust and better debuggable. Usually you will
* never want to change this.
*/
@NotNull
private Duration genericMediumDelay = Duration.ofSeconds(5);
/**
* By default, how long should we expect a task to remain in any state, before we consider it as stuck.
*
* Notice, that it is not used for PROCESSING state, where the maximum time is asked from task handler itself.
*/
@NotNull
private Duration taskStuckTimeout = Duration.ofMinutes(30);
/**
* How much do we load triggers from triggering topic into memory, aka look-ahead amount.
*/
@Min(1L)
private int maxTriggersInMemory = 100000;
/**
* How many triggers maximum do we retrieve from Kafka with one polling loop.
*/
@Min(1L)
private int triggerFetchSize = 100;
/**
* How many nodes do we expect to be in the cluster.
*/
private int maxNodeCount = 2;
/**
* We have two triggering algorithms. First one spreads task triggerings into partitions and every service node is listening only its own partition.
* The cons is that if one node gets very slow, some tasks latency goes up, even when other nodes would have processing power for them. In second
* algorithms every node is taking every trigger from topic and tries to execute them. The cons is that it may be inefficient with large cluster
* size (10+), even when we actually use very efficient optimistic locking for a node to grab a task for itself.
*
*
It does not work well in cluster, where node ids change. Like, for example, in kubernetes. So if service in kubernetes don't set it to true.
*
*
So rule of thumb is that you probably want to use the second algorithm for cluster with less than 10 nodes.
*
*
Second algorithm is activated with setting this parameter to true.
*/
private boolean triggerSameTaskInAllNodes = false;
/**
* Connection string to Zookeeper. Used to set partition sizes for different topics.
*/
@NotBlank
@ResolvedValue
private String zookeeperConnectString;
/**
* Topic replication factor for listened topics and task triggering topics.
*/
private short topicReplicationFactor = 3;
/**
* MySQL or Postgres.
*/
@NotNull
private DbType dbType;
/**
* MDC keys config.
*/
@Valid
@NotNull
private Mdc mdc = new Mdc();
/**
* We support Transferwise Kafka failover, where for every topic, we additionally listen to 2 other topics, one starting with "fra." and other with
* ".aws.".
*
*
e.g. kafkaDataCenterPrefixes = "fra.,aws.";
*/
@ResolvedValue
private String kafkaDataCenterPrefixes = "";
/**
* Sometimes environments and engineers are forced to use same Kafka server, but still want to deal with only their own messages. In that case we
* can configure a so called namespace string, which is prepended to every topics name.
*/
@ResolvedValue
private String kafkaTopicsNamespace;
/**
* Set it to false, if you use proper transaction manager and not spring's default one. You will get better performance and waste less memory.
* Default option true prevents possible deadlocks with any transaction manager.
*/
private boolean asyncTaskTriggering = true;
/**
* Tied to the previous option. If asyncTaskTriggering is enabled, how many triggerings to we keep in memory, before starting to throttle new tasks
* added.
*/
@Min(1L)
private int maxAsyncTaskTriggerings = 100000;
/**
* In how many threads to we try to trigger tasks when using crappy Spring own transaction manager.
*/
@Min(1L)
private int asyncTaskTriggeringsConcurrency = 10;
/**
* Highest task priority allowed.
*/
private int highestPriority = 0;
/**
* Lowest task priority allowed.
*/
private int lowestPriority = 9;
/**
* When we lose the offset of a triggering topic, where do we rewind? Only used for task triggering. For usual topics listeners, the spring-kafka
* configuration is used.
*
*
Can use "earliest", "latest" or Duration notion. For example, if you want to rewind 30 min back, you should write "-PT30M";
*/
@NotBlank
@ResolvedValue
private String autoResetOffsetTo = "-PT30M";
/**
* When do we consider a task or task unique key old enough to be removed from the database.
*/
@NotNull
private Duration finishedTasksHistoryToKeep = Duration.ofDays(30);
/**
* How many old tasks maximum do we delete in one batch/transaction. Deletion should always happen in small batches to not create too big spikes for
* database replication.
*
*
This can handle 10 tasks/s.
*
*
TODO: Implement dynamic, adaptive configuration/system for that instead. Batch Size could be constant, but interval should learn from current
* situation. Can use TCP/IP flow control algorithms.
*/
@Min(1L)
private int tasksHistoryDeletingBatchSize = 2 * 125;
//TODO: This does not make sense as generic parameter.
// taskhandler should provide this info programmatically.
// No usage so far in Tw
/**
* Should we delete a task immediately after it has marked as DONE. Maybe in the future the task handler can provide this information.
*
*
The value of this property depends on balance between keeping the storage used minimal or being able to later track,
* analyze or even force-retry executed tasks.
*
*
Notice, that if you set it to true, you currently lose the taskId based uniqueness checks.
*/
private boolean deleteTaskOnFinish = false;
/**
* Removes the payload but keeps task record in database. Useful for huge payloads where uniqueness checks are still desired.
*/
private boolean clearPayloadOnFinish = false;
/**
* How long a task has to be stuck, before we start sending out VictorOps alerts.
*/
@NotNull
private Duration stuckTaskAge = Duration.ofMinutes(5);
private boolean checkVersionBeforeGrabbing = false;
private boolean assertStatusOnGrabbing = false;
/**
* The additional task buckets, not including the default bucket, that we will process.
*
*
If a task handler is configured with a bucket not present in this list, then the handler will not be invoked when new tasks of the configured
* type are submitted, and instead the task will be sent to the error state.
*/
@NotNull
private List additionalProcessingBuckets = new ArrayList<>();
/**
* Need to make it configurable as in some environments, like smoke tests, we don't need a zookeeper connection.
*/
private boolean preventStartWithoutZookeeper = true;
@NotBlank
@ResolvedValue
private String taskTableName = "tw_task";
@NotBlank
@ResolvedValue
private String uniqueTaskKeyTableName = "unique_tw_task_key";
@NotBlank
@ResolvedValue
private String taskDataTableName = "tw_task_data";
@NotNull
@ResolvedValue
private String taskTablesSchemaName = "";
/**
* For extremely latency sensitive scenarios or for tests, we allow to trigger directly in the same process, instead of going through the kafka
* pipes. TODO: Maybe allow to execute service tests without having no Kafka at all. Probably best to have a separate ITasksExecutionTrigger
* implementation instead of hacking it into Kafka one.
*/
private boolean triggerInSameProcess;
/**
* Just to allow `ignoreUnknownFields` work.
*/
@ResolvedValue
private String baseUrl;
/**
* Safety limit, to not kill database performance, when something goes horribly wrong. For example when we have millions of waiting, erronous or
* stuck tasks.
*
* The side effect is, that for example erroneous tasks count will never exceed this number.
*/
@Min(1)
private int maxDatabaseFetchSize = 10000;
/**
* Allows to turn off automatic start of tasks processing. In technical terms, allows to turn off fetching of task triggers and processing those.
*
*
Does not apply when tasks are triggered with `triggerInSameProcess` system.
*/
private boolean autoStartProcessing = true;
/**
* Experimental, do not use.
*/
private Duration interruptTasksAfterShutdownTime = null;
/**
* Adds more counters showing more details for the processing engine.
*
*
Adds a considerable overhead and interpreting results needs deep understanding of tw-tasks code.
*
*
Meant to be used only by tw-tasks contributors when helping to solve some very specific incident.
*/
private boolean debugMetricsEnabled = false;
/**
* Code is running some assertions.
*
*
Only meant to be true in tw-tasks own test-suite.
*/
private boolean assertionsEnabled = false;
/**
* If true, the task cleaning will also handle those cases consistently where just-to-be deleted tasks may change.
*
*
It makes the cleaning process a bit less efficient and it is almost never needed.
*/
private boolean paranoidTasksCleaning = false;
/**
* How many tasks per bucket we are trying to grab at the same time.
*
*
Mainly meant as a soft safety measure in cases where concurrency policies are lacking good quality.
*
*
The amount of tasks grabbings happening at the same time, is also limited by the concurrency policies.
*
*
The higher the latency between application and the database, the more useful a larger number can be.
*
*
The default 25 is somewhat optimized for RDS Multi A/Z databases with high commit latency, where we have 6 nodes application cluster,
* relatively close to the database.
*/
@Min(1)
private Integer taskGrabbingMaxConcurrency = 25;
/**
* Cluster wide tasks state monitoring options.
*/
@Valid
private ClusterWideTasksStateMonitor clusterWideTasksStateMonitor = new ClusterWideTasksStateMonitor();
public enum DbType {
MYSQL, POSTGRES
}
@Valid
private TasksManagement tasksManagement = new TasksManagement();
@Valid
private Compression compression = new Compression();
@Valid
private Environment environment = new Environment();
@Valid
private Triggering triggering = new Triggering();
@Valid
private TasksResumer tasksResumer = new TasksResumer();
@Data
public static class Triggering {
@Valid
private Kafka kafka = new Kafka();
@Data
public static class Kafka {
@NotBlank
@ResolvedValue
private String bootstrapServers;
/**
* Allows to override configuration properties for both Kafka Consumers. and Producers.
*/
private Map properties = new HashMap<>();
}
}
@Data
public static class TasksManagement {
/**
* A role for viewing PII data.
*/
@NotNull
private Set viewTaskDataRoles = new HashSet<>(Collections.singletonList("NONEXISTING_ROLE_FOR_TESTING_PURPOSES_ONLY"));
/**
* Roles for all other task management endpoints.
*/
@NotNull
private Set roles = new HashSet<>(Collections.singleton("ROLE_DEVEL"));
@NotNull
@Valid
private List typeSpecific = Collections.emptyList();
@Data
public static class TypeSpecificTaskManagement {
@NotBlank
@ResolvedValue
private String taskType;
@NotEmpty
private Set viewTaskDataRoles = new HashSet<>(Collections.singletonList("NONEXISTING_ROLE_FOR_TESTING_PURPOSES_ONLY"));
}
}
/**
* Allows to specify MDC keys used.
*/
@Data
public static class Mdc {
@NotBlank
@ResolvedValue
private String taskIdKey = "twTaskId";
@NotBlank
@ResolvedValue
private String taskVersionKey = "twTaskVersion";
@NotBlank
@ResolvedValue
private String taskTypeKey = "twTaskType";
@NotBlank
@ResolvedValue
private String taskSubTypeKey = "twTaskSubType";
}
/**
* Cluster-wide monitoring config.
*/
@Data
public static class ClusterWideTasksStateMonitor {
/**
* How often does the monitor approximately run.
*
* Monitor can actually run slower or faster, when leadership is switching rapidly.
*/
@NotNull
private Duration interval = Duration.ofSeconds(30);
/**
* The time between monitor acquires leadership and first check is done.
*/
@NotNull
private Duration startDelay = Duration.ofSeconds(5);
/**
* If enabled, we will gather approximate tasks and unique keys counts from database information schema tables.
*/
private boolean tasksCountingEnabled = true;
}
@Data
@Accessors(chain = true)
public static class Compression {
@NotNull
private CompressionAlgorithm algorithm = CompressionAlgorithm.GZIP;
/**
* Can be quite large, even when we have small(er) messages, because we reuse memory buffers.
*/
private Integer blockSizeBytes;
/**
* Approximate message size is considered.
*/
private int minSize = 128;
/**
* Used when applicable.
*/
private Integer level;
}
@Data
@Accessors(chain = true)
public static class Environment {
/**
* Version deployed (e.g. to production).
*
*
Allows tw-tasks to decide when it should fail fast, instead of risking with incompatibilities or/and processing pauses.
*/
@NotBlank
@ResolvedValue
private String previousVersion;
}
@Data
@Accessors(chain = true)
public static class TasksResumer {
/**
* Specifies how many tasks we are loading from the database in one go to be then resumed concurrently.
*/
@Positive
private int batchSize = 1000;
@Positive
private int concurrency = 10;
}
}