# Maximum number of concurrent requests to accept in a single
# service process, or 0 for no limit.
- MaxConcurrentRequests: 0
+ MaxConcurrentRequests: 64
+
+ # Fraction of MaxConcurrentRequests that can be "log create"
+ # messages at any given time. This is to prevent logging
+ # updates from crowding out more important requests.
+ LogCreateRequestFraction: 0.50
# Maximum number of 64MiB memory buffers per Keepstore server process, or
# 0 for no limit. When this limit is reached, up to
# any user with "manage" permission can un-freeze.
UnfreezeProjectRequiresAdmin: false
+ # (Experimental) Use row-level locking on update API calls.
+ LockBeforeUpdate: false
+
Users:
# Config parameters to automatically setup new users. If enabled,
# this users will be able to self-activate. Enable this if you want
# params_truncated.
MaxRequestLogParamsSize: 2000
+ # In all services except RailsAPI, periodically check whether
+ # the incoming HTTP request queue is nearly full (see
+ # MaxConcurrentRequests) and, if so, write a snapshot of the
+ # request queue to {service}-requests.json in the specified
+ # directory.
+ #
+ # Leave blank to disable.
+ RequestQueueDumpDirectory: ""
+
Collections:
# Enable access controls for data stored in Keep. This should
#
# If SIGUSR1 is received during an idle period between operations,
# the next operation will start immediately.
- BalancePeriod: 10m
+ BalancePeriod: 6h
# Limits the number of collections retrieved by keep-balance per
# API transaction. If this is zero, page size is
BalanceCollectionBatch: 0
# The size of keep-balance's internal queue of
- # collections. Higher values use more memory and improve throughput
- # by allowing keep-balance to fetch the next page of collections
- # while the current page is still being processed. If this is zero
- # or omitted, pages are processed serially.
- BalanceCollectionBuffers: 1000
+ # collections. Higher values may improve throughput by allowing
+ # keep-balance to fetch collections from the database while the
+ # current collection are still being processed, at the expense of
+ # using more memory. If this is zero or omitted, pages are
+ # processed serially.
+ BalanceCollectionBuffers: 4
# Maximum time for a rebalancing run. This ensures keep-balance
# eventually gives up and retries if, for example, a network
# Skip TLS certificate name verification.
InsecureTLS: false
+ # Mininum TLS version to negotiate when connecting to server
+ # (ldaps://... or StartTLS). It may be necessary to set this
+ # to "1.1" for compatibility with older LDAP servers that fail
+ # with 'LDAP Result Code 200 "Network Error": TLS handshake
+ # failed (tls: server selected unsupported protocol version
+ # 301)'.
+ #
+ # If blank, use the recommended minimum version (1.2).
+ MinTLSVersion: ""
+
# Strip the @domain part if a user supplies an email-style
# username with this domain. If "*", strip any user-provided
# domain. If "", never strip the domain part. Example:
# probably want to include the other Workbench instances in the
# federation in this list.
#
+ # A wildcard like "https://*.example" will match client URLs
+ # like "https://a.example" and "https://a.b.c.example".
+ #
# Example:
#
# TrustedClients:
# disk cache size will use a disk cache, sized to the
# container's RAM requirement (but with minimum 2 GiB and
# maximum 32 GiB).
+ #
+ # Note: If you change this value, containers that used the previous
+ # default value will only be reused by container requests that
+ # explicitly specify the previous value in their keep_cache_ram
+ # runtime constraint.
DefaultKeepCacheRAM: 0
# Number of times a container can be unlocked before being
# with the cancelled container.
MaxRetryAttempts: 3
- # The maximum number of compute nodes that can be in use simultaneously
- # If this limit is reduced, any existing nodes with slot number >= new limit
- # will not be counted against the new limit. In other words, the new limit
- # won't be strictly enforced until those nodes with higher slot numbers
- # go down.
- MaxComputeVMs: 64
-
# Schedule all child containers on preemptible instances (e.g. AWS
# Spot Instances) even if not requested by the submitter.
#
# Maximum bytes that may be logged by a single job. Log bytes that are
# silenced by throttling are not counted against this total.
+ # If you set this to zero, each container will only create a single
+ # log on the API server, noting for users that logging is throttled.
LimitLogBytesPerJob: 67108864
LogPartialLineThrottlePeriod: 5s
# providers too, if desired.
MaxConcurrentInstanceCreateOps: 1
+ # The maximum number of instances to run at a time, or 0 for
+ # unlimited.
+ #
+ # If more instances than this are already running and busy
+ # when the dispatcher starts up, the running containers will
+ # be allowed to finish before the excess instances are shut
+ # down.
+ MaxInstances: 64
+
+ # Maximum fraction of CloudVMs.MaxInstances allowed to run
+ # "supervisor" containers at any given time. A supervisor is a
+ # container whose purpose is mainly to submit and manage other
+ # containers, such as arvados-cwl-runner workflow runner.
+ #
+ # If there is a hard limit on the amount of concurrent
+ # containers that the cluster can run, it is important to
+ # avoid crowding out the containers doing useful work with
+ # containers who just create more work.
+ #
+ # For example, with the default MaxInstances of 64, it will
+ # schedule at most floor(64*0.30) = 19 concurrent workflows,
+ # ensuring 45 slots are available for work.
+ SupervisorFraction: 0.30
+
# Interval between cloud provider syncs/updates ("list all
# instances").
SyncInterval: 1m
# https://xxxxx.blob.core.windows.net/system/Microsoft.Compute/Images/images/xxxxx.vhd
ImageID: ""
+ # Shell script to run on new instances using the cloud
+ # provider's UserData (EC2) or CustomData (Azure) feature.
+ #
+ # It is not necessary to include a #!/bin/sh line.
+ InstanceInitCommand: ""
+
# An executable file (located on the dispatcher host) to be
# copied to cloud instances at runtime and used as the
# container runner/supervisor. The default value is the
ReadTimeout: 10m
RaceWindow: 24h
PrefixLength: 0
- # Use aws-s3-go (v2) instead of goamz
- UseAWSS3v2Driver: true
# For S3 driver, potentially unsafe tuning parameter,
# intentionally excluded from main documentation.
# This feature is disabled when set to zero.
IdleTimeout: 0s
- # URL to a file that is a fragment of text or HTML which should
- # be rendered in Workbench as a banner.
- BannerURL: ""
+ # UUID of a collection. This collection should be shared with
+ # all users. Workbench will look for a file "banner.html" in
+ # this collection and display its contents (should be
+ # HTML-formatted text) when users first log in to Workbench.
+ BannerUUID: ""
# Workbench welcome screen, this is HTML text that will be
# incorporated directly onto the page.
projects / arvados.git / blobdiff