Singularity Configuration

Singularity (Service) is configured by DropWizard via a YAML file referenced on the command line. Top-level configuration elements reside at the root of the configuration file alongside DropWizard configuration.

Root Configuration

Common Configuration

These are settings that are more likely to be altered.


Parameter Default Description Type
allowRequestsWithoutOwners true If false, submitting a request without at least one owner will return a 400 boolean
commonHostnameSuffixToOmit null If specified, will remove this hostname suffix from all taskIds string
defaultSlavePlacement GREEDY See Agent Placement enum / string [GREEDY, OPTIMISTIC, SEPARATE (deprecated), SEPARATE_BY_DEPLOY, SEPARATE_BY_REQUEST, SPREAD_ALL_SLAVES]
defaultValueForKillTasksOfPausedRequests true When a task is paused, the API allows for the tasks of that request to optionally not be killed. If that parameter is not set in the pause request, this value is used boolean
deltaAfterWhichTasksAreLateMillis 30000 (30 seconds) The amount of time after a task's schedule time that Singularity will classify it (in state API and dashboard) as a late task long
deployHealthyBySeconds 120 Default amount of time to allow pending deploys to run for before transitioning them into active deploys. If more than this time passes before a deploy can be considered healthy (all of its tasks either make it to TASK_RUNNING or pass healthchecks), then the deploy will be rejected long
killNonLongRunningTasksInCleanupAfterSeconds 86400 (1 day) Kills scheduled and one-off tasks after this amount of time if they have been scheduled for cleaning (a new deploy succeeds, the underlying agent is decomissioned) long
hostname null Hostname of this Singularity instance string

Healthchecks and New Task Checks

Parameter Default Description Type
considerTaskHealthyAfterRunningForSeconds 5 Tasks which make it to TASK_RUNNING and run for at least this long (that are not health-checked) are considered healthy long
healthcheckIntervalSeconds 5 Default amount of time to wait in between attempting task healthchecks int
healthcheckTimeoutSeconds 5 Default amount of time to wait for healthchecks to return before considering them failed int
killAfterTasksDoNotRunDefaultSeconds 600 (10 minutes) Amount of time after which new tasks (that are not part of a deploy) will be killed if they do not enter TASK_RUNNING long
healthcheckMaxRetries Default max number of time to retry a failed healthcheck for a task before considering the task to be unhealthy int
startupDelaySeconds By default, wait this long before starting any healthchecks on a task int
startupTimeoutSeconds 45 If a healthchecked task has not responded with a valid http response in startupTimeoutSeconds consider it unhealthy int
startupIntervalSeconds 2 In the startup period (before a valid http response has been received) wait this long between healthcheck attempts int
healthcheckFailureStatusCodes [] If any of these status codes is received during a healthcheck, immediately consider the task unhealthy, do not retry the check List


Parameter Default Description Type
defaultDeployStepWaitTimeMs 0 If using an incremental deploy, wait this long between deploy steps if not specified in the deploy int
defaultDeployMaxTaskRetries 0 Allow this many tasks to fail and be retried before failing a new deploy int
allowDeployOfPausedRequests false If true, paused requests can be deployed without unpausing or starting new tasks at deploy time boolean


Parameter Default Description Type
maxDeployIdSize 50 Deploy ids over this size will cause deploy requests to fail with 400 int
maxRequestIdSize 100 Request ids over this size will cause new requests to fail with 400 int


Cooldown is divided into 2 types, fast and slow. These are essentially two sets of differing thresholds for cooldown, meant to act quickly for cases where there are rapid failures, but still provide a notification/signal for cases where there are slow but repeated failures

Parameter Default Description Type
fastFailureCooldownCount/slowFailureCooldownCount 3/5 The number of sequential failures after which a request is placed into system cooldown int
fastFailureCooldownMs/slowFailureCooldownMs 30000/600000 The time window during which ...CooldownCount failures must occur long
fastCooldownExpiresMinutesWithoutFailure/slowCooldownExpiresMinutesWithoutFailure 5/5 If there are no failures after this time period, the request will exit cooldown int
cooldownMinScheduleSeconds 120 When a request enters cooldown, new tasks are delayed by at least this long long

Load Balancer API

Parameter Default Description Type
loadBalancerQueryParams null Additional query parameters to pass to the Load Balancer API Map
loadBalancerRequestTimeoutMillis 2000 The timeout for making API calls to the Load Balancer API (these will be retried) long
loadBalancerUri null The URI of the Load Balancer API (Baragon) string
deleteRemovedRequestsFromLoadBalancer false If a request is removed from Singularity, issue a DELETE to the load balancer for that service boolean

User Interface

Parameter Default Description Type
sandboxDefaultsToTaskId false If true, the Singularity API will return the sandbox view of root/taskId when queried without a path (Useful when using SingularityExecutor) boolean
enableCorsFilter false If true, provides a Bundle which will enable CORS boolean

Internal Scheduler Configuration

These settings are less likely to be changed, but were included in the configuration instead of hardcoding values.


Parameter Default Description Type
checkDeploysEverySeconds 5 Check the status (health) of pending deploys, promoting them to active or removing them on this interval long
checkNewTasksEverySeconds 5 Check the health of new (non-deployed, non-healthchecked) tasks to make sure they eventually get to running on this interval long
checkSchedulerEverySeconds 5 Runs scheduler checks (processes decommissions and pending queue) on this interval (these tasks also run when an offer is received) long
checkWebhooksEveryMillis 10000 (10 seconds) Will check for and send new queued webhooks on this interval long
cleanupEverySeconds 5 Will cleanup request, task, and other queues on this interval long
persistHistoryEverySeconds 3600 (1 hour) Moves stale historical task data from ZooKeeper into the database, setting to 0 will disable history persistence long
saveStateEverySeconds 60 State about this Singularity instance is saved (available over API) on this interval long
checkJobsEveryMillis 600000 (10 mins) Check for jobs running longer than the expected time on this interval long
checkExpiringUserActionEveryMillis 45000 Check for expiring actions that should be expired on this interval long


Parameter Default Description Type
checkReconcileWhenRunningEveryMillis 30000 (30 seconds) When reconciling tasks, will re-request task updates on this interval until reconciliation finishes long
startNewReconcileEverySeconds 600 (10 minutes) Starts a new reconciliation cycle (if one is not currently running) on this interval (A relatively costly operation that detects updates Mesos failed to deliver) long
askDriverToKillTasksAgainAfterMillis 300000 (5 minutes) Amount of time to wait before instruction mesos to kill a task which has been killed by Singularity but is still running long

Thread Pools

Parameter Default Description Type
checkNewTasksScheduledThreads 3 Max number of threads to use to check new tasks int
healthcheckStartThreads 3 Max number of threads to use to start healthchecks int
logFetchMaxThreads 15 Max number of threads to use to fetch log directories from Mesos REST API int


Parameter Default Description Type
closeWaitSeconds 5 Will wait at least this many seconds when shutting down thread pools long
compressLargeDataObjects true Will compress larger objects inside of ZooKeeper and the database boolean
maxHealthcheckResponseBodyBytes 8192 Number of bytes to save from healthcheck responses (displayed in UI) int
maxQueuedUpdatesPerWebhook 50 Max number of updates to queue for a given webhook url, after which some webhooks will not be delivered int
zookeeperAsyncTimeout 5000 Milliseconds for ZooKeeper timeout. Calls to ZooKeeper which take over this timeout will cause the operations to fail and Singularity to abort long
cacheStateForMillis 30000 (30 seconds) Amount of time to cache internal state for when requested over API long
sandboxHttpTimeoutMillis 5000 (5 seconds) Sandbox HTTP calls will timeout after this amount of time (fetching logs for emails / UI)
newTaskCheckerBaseDelaySeconds 1 Added to the the amount of deploy to wait before checking a new task long
allowTestResourceCalls false If true, allows calls to be made to the test resource, which can test internal methods boolean
deleteDeploysFromZkWhenNoDatabaseAfterHours 336 (14 days) Delete deploys from zk when they are older than this if we are not using a database long
maxStaleDeploysPerRequestInZkWhenNoDatabase infinite (disabled) Delete oldest deploys from zk when there are more than this number for a given request, if we're not already persisting them to a database int
deleteStaleRequestsFromZkWhenNoDatabaseAfterHours 336 (14 days) Delete stale requests after this amount of time if we are not using a database long
maxRequestsWithHistoryInZkWhenNoDatabase infinite (disabled) Delete history of oldest requests from zk when there are more than this number of requests, if we're not already persisting them to a database int
deleteTasksFromZkWhenNoDatabaseAfterHours 168 (7 days) Delete old tasks from zk after this amount of time if we are not using a database long
maxStaleTasksPerRequestInZkWhenNoDatabase infinite (disabled) Delete oldest tasks from zk when there are more than this number for a given request, if we're not already persisting them to a database int
taskPersistAfterStartupBufferMillis 60000ms (1 min) Wait this long after a task starts before persisting it in history long
deleteDeadSlavesAfterHours 168 (7 days) Remove dead agents from the list after this amount of time long
deleteUndeliverableWebhooksAfterHours 168 (7 days) Delete (and stop retrying) failed webhooks after this amount of time long
waitForListeners true If true, the event system waits for all listeners having processed an event. boolean
warnIfScheduledJobIsRunningForAtLeastMillis 86400000 (1 day) Warn if a scheduled job has been running for this long long
warnIfScheduledJobIsRunningPastNextRunPct 200 Warn if a scheduled job has run this much past its next scheduled run time (e.g. 200 => ran through next two run times) int
pendingDeployHoldTaskDuringDecommissionMillis 600000ms (10 minutes) Don't kill tasks on a decommissioning agent that are part of a pending deploy for this amount of time to allow the deploy to complete long
defaultBounceExpirationMinutes 60 Expire a bounce after this many minutes if an expiration is not provided in the request to bounce int
cacheOffers false Hold on to unused offers for up to cacheOffersForMillis boolean
cacheOffersForMillis If cacheOffers is true, decline offers after this amount of time if they ahve not been used long
offerCacheSize The maximum number of offers to cache at once int

Mesos Configuration

These settings should live under the "mesos" field inside the root configuration.


Parameter Default Description Type
master null A comma separated list of mesos master http(s)://user:password@host:port user and password are optional, http is used if no protocol is provided String
frameworkName null String
frameworkId null String
frameworkFailoverTimeout 0.0 double
frameworkRole null Specify framework's desired role when Singularity registers with the master String
checkpoint true boolean
credentialPrincipal Used to enable authorization based on the authenticated principal String

Resource Limits

Parameter Default Description Type
defaultCpus 1 Number of CPUs to request for a task if none are specified int
defaultMemory 64 MB of memory to request for a task if none is specified int
defaultDisk 1024 MB of disk to request for a task if none is specified int
maxNumInstancesPerRequest 25 Max instances (tasks) to allow for a request (requests using over this will return a 400) int
maxNumCpusPerInstance 50 Max number of CPUs allowed on a given task int
maxNumCpusPerRequest 900 Max number of CPUs allowed for a given request (cpus per task * task instance) int
maxMemoryMbPerInstance 24000 Max MB of memory allowed on a given task int
maxMemoryMbPerRequest 450000 Max MB of memory allowed for a given request (memoryMb per task * task instances) int


Parameter Default Description Type
rackIdAttributeKey rackid The Mesos agent attribute to denote a rack string
defaultRackId DEFAULT The rackId to assign to a agent if no rackId attribute value is present string


Parameter Default Description Type
agentHttpPort 5051 The port to talk to agents on int
agentHttpsPort absent The HTTPS port to talk to agents on Integer (Optional)


Parameter Default Description Type
allocatedResourceWeight 0.5 This portion of an offer's score depends on the amount of resources currently allocated by mesos on the mesos agent double
inUseResourceWeight 0.5 This portion of an offer's score depends on the currently used resources on a mesos agent as reported by the agent statistics endpoint double
cpuWeight 0.4 The weight the agent's cpu carries when scoring an offer double
memWeight 0.4 The weight the agent's memory carries when scoring an offer double
diskWeight 0.2 The weight the agent's disk carries when scoring an offer double


Parameter Default Description Type
database The database connection for SingularityService follows the dropwizard DataSourceFactory format DataSourceFactory

Network Configuration

These settings should live under the "network" field of the root configuration.

Parameter Default Description Type
defaultPortMapping false If no port mapping is provided, map all Mesos-provided ports to the host boolean

History Purging

These settings live under the "historyPuring" field in the root configuration

Parameter Default Description Type
deleteTaskHistoryAfterDays 365 Purge tasks older than this many days int
deleteTaskHistoryAfterTasksPerRequest 10000 Purge oldest tasks when there are more than this many associated with a single request int
deleteTaskHistoryBytesInsteadOfEntireRow true Only delete the taskHistoryBytes instead of the entire record of the task (e.g. to save space) boolean
checkTaskHistoryEveryHours 24 Run the purge every x hours int
enabled false Should we run the database purge boolean


These settings live under the "s3" field in the root configuration. If using the SingularityS3Uploader, this section will need to be provided in order to view lists of and download s3 logs from the SingularityUI.

Parameter Default Description Type
maxS3Thread 3 Max threads to run for fetching logs from s3 int
waitForS3ListSeconds 5 Timeout in seconds for fetching list of s3 logs int
waitForS3LinksSeconds 1 Timeout in seconds for creating new s3 links int
expireS3LinksAfterMillis 86400000 (1 day) Expire generated s3 log links after this amount of time long
s3Bucket S3 bucket to search for logs String
groupOverrides Extra s3 configurations provided such that individual requests may use separate s3 buckets. Each S3GroupOverrideConfiguration has a name specified by the Map key and consists of an s3Bueckt, s3AccessKey, and s3SecretKey Map
s3KeyFormat Search for logs with keys in this format, should be the same as the key format set in the SingularityS3Uploader String
s3AccessKey aws access key for the specified s3 bucket String
s3SecretKey aws secret key for the specified s3 bucket String
missingTaskDefaultS3SearchPeriodMillis 259200000ms (3 days) Search over this many days for s3 logs when no task data is found long


These settings live under the "sentry" field in the root config and enable Singularity error reporting to sentry.

Parameter Default Description Type
dsn Sentry DSN (Data Source Name) String
prefix "" Prefix string for event culprit naming and messages String


These settings live under the "smtp" field in the root config.

Parameter Default Description Type
username smtp username String
password smtp password String
taskLogLength 512 Send this many lines of a tasks log in emails int
host localhost Host for smtp session String
port 25 Port for smtp session int
from "" Send emails form this address String
mailMaxThreads 3 max threads for email sending process int
admins [] List of admin user emails List\
rateLimitAfterNotifications 5 Rate limit email sending after this many notifications have been sent in rateLimitPeriodMillis int
rateLimitPeriodMillis 60000 (10 mins) time period for rateLimitAfterNotifications long
rateLimitCooldownMillis 3600000 (1 hour) Cooldown time before rate limiting is removed long
taskEmailTailFiles [stdout, stderr] Send the tail of these files in messages about tasks List\
emails See below See below Map\>
subjectPrefix unset String prepended to the email subject line String
ssl false Connect to SMTP host over ssl boolean

You may need libmail-java installed on your Singularity master host in order to connect to your smtp server.

Emails List

The emails list determines what emails to send notifications to and for what events. You can specify a map of EmailType to a list of EmailDestinations

EmailType corresponds to different events that could trigger emails such as TASK_LOST or TASK_FAILED

EmailDestination corresponds to one of OWNERS (as listed on the Singularity Request), ACTION_TAKER (user who triggered the action causing the email update), or ADMINS (specified in config as seen above)

An email list might look something like

      - OWNERS
      - OWNERS
      - OWNERS
      - OWNERS
      - OWNERS
      - OWNERS
      - OWNERS
      - OWNERS
      - OWNERS
      - OWNERS
      - OWNERS

UI Configuration

These settings live under the "ui" field in the root config.

Parameter Default Description Type
title "Singularity" Title shown in the left of the menu bar in ui String
navColor "" Color for nav bar String
baseUrl Base url where the ui will be hosted (e.g. http://localhost:7099/singularity) String
runningTaskLogPath stdout Generate link to this log for running tasks on the request page String
finishedTaskLogPath stdout Generate link to this log for finished tasks on the request page String
hideNewDeployButton false Don't show the 'New Deploy' button boolean
hideNewRequestButton false Don't show the 'New Request' button boolean
rootUrlMode INDEX_CATCHALL INDEX_CATCHALL: UI is served off of / using a catchall resource. UI_REDIRECT: UI is served off of /ui, path and index redirects there. DISABLED: UI is served off of /ui and the root resource is not served at all enum / String INDEX_CATCHALL, UI_REDIRECT, DISABLED


These settings live under the "zookeeper" field in the root config.

Parameter Default Description Type
quorum Comma separated host:port list of zk hosts String
sessionTimeoutMillis 600_000 zookeeper session timeout int
connectTimeoutMillis 60_000 Connect to zookeeper timeout int
retryBaseSleepTimeMilliseconds 1_000 Wait time between zookeeper connection retries int
retryMaxTries 3 Max retries to obtain a zookeeper connection before aborting int
zkNamespace Path under which to store Singularity data in zk (e.g. /singularity) String