disco.settings – Disco Settings

Settings can be specified in a Python file and/or using environment variables. Settings specified in environment variables override those stored in a file. The default settings are intended to make it easy to get Disco running on a single node. make install will create a more reasonable settings file for a cluster environment, and put it in /etc/disco/settings.py

Disco looks in the following places for a settings file:

  • The settings file specified using the command line utility --settings option.
  • ~/.disco
  • /etc/disco/settings.py

Possible settings for Disco are as follows:

DISCO_DATA

Directory to use for writing data. Default obtained using os.path.join(DISCO_ROOT, data).

DISCO_DEBUG

Sets the debugging level for Disco. Default is 1.

DISCO_ERLANG

Command used to launch Erlang on all nodes in the cluster. Default usually erl, but depends on the OS.

DISCO_EVENTS

If set, events are logged to stdout. If set to json, events will be written as JSON strings. If set to nocolor, ANSI color escape sequences will not be used, even if the terminal supports it. Default is unset (the empty string).

DISCO_FLAGS

Default is the empty string.

DISCO_HOME

The directory which Disco runs out of. If you run Disco out of the source directory, you shouldn’t need to change this. If you use make install to install Disco, it will be set properly for you in /etc/disco/settings.py.

DISCO_HTTPD

Command used to launch lighttpd. Default is lighttpd.

DISCO_MASTER_HOME

Directory containing the Disco master directory. Default is obtained using os.path.join(DISCO_HOME, 'master').

DISCO_MASTER_HOST

The hostname of the master. Default obtained using socket.gethostname().

DISCO_MASTER_ROOT

Directory to use for writing master data. Default obtained using os.path.join(DISCO_DATA, '_%s' % DISCO_NAME).

DISCO_MASTER_CONFIG

Directory to use for writing cluster configuration. Default obtained using os.path.join(DISCO_ROOT, '%s.config' % DISCO_NAME).

DISCO_NAME

A unique name for the Disco cluster. Default obtained using 'disco_%s' % DISCO_PORT.

DISCO_LOG_DIR

Directory where log-files are created. The same path is used for all nodes in the cluster. Default is obtained using os.path.join(DISCO_ROOT, 'log').

DISCO_PID_DIR

Directory where pid-files are created. The same path is used for all nodes in the cluster. Default is obtained using os.path.join(DISCO_ROOT, 'run').

DISCO_PORT

The port the workers use for HTTP communication. Default is 8989.

DISCO_ROOT

Root directory for Disco-written data and metadata. Default is obtained using os.path.join(DISCO_HOME, 'root').

DISCO_ROTATE_LOG

Whether to rotate the master log on startup. Default is False.

DISCO_USER

The user Disco should run as. Default obtained using os.getenv(LOGNAME).

DISCO_JOB_OWNER

User name shown on the job status page for the user who submitted the job. Default is the login name @ host.

DISCO_WWW_ROOT

Directory that is the document root for the master HTTP server. Default obtained using os.path.join(DISCO_MASTER_HOME, www).

DISCO_GC_AFTER

How long to wait before garbage collecting job-generated intermediate and result data. Only results explictly saved to DDFS won’t be garbage collected. Default is 100 * 365 * 24 * 60 * 60 (100 years). (Note that this setting does not affect data in DDFS.)

DISCO_PROFILE

Whether Disco should start profiling applications and send profiling data to a graphite server.

GRAPHITE_HOST
If DISCO_PROFILE is set, then some performance data from Disco
will be sent to the graphite host. The default is localhost.
We are assuming that the listening port is the default graphite
port.
SYSTEMD_ENABLED

This adds -noshell to the erlang process. It provides compatibility for running disco using a non-forking process type in the service definition.

DISCO_WORKER_MAX_MEM

How much memory can be used by worker in total. Worker calls resource.setrlimit(RLIMIT_AS, limit) to set the limit when it starts. Can be either a percentage of total available memory or an exact number of bytes. Note that setrlimit behaves differently on Linux and Mac OS X, see man setrlimit for more information. Default is 80% i.e. 80% of the total available memory.

Settings to control the proxying behavior:

DISCO_PROXY_ENABLED

If set, enable proxying through the master. This is a master-side setting (set in master:/etc/disco/settings.py). Default is ''.

DISCO_PROXY

The address of the proxy to use on the client side. This is in the format http://<proxy-host>:<proxy-port>, where <proxy-port> normally matches the value of DISCO_PROXY_PORT set on the master.

Default is ''.

DISCO_PROXY_PORT

The port the master proxy should run on. This is master-side setting (set in master:/etc/disco/settings.py). Default is 8999.

Settings to control the scheduler behavior:

DISCO_SCHEDULER

The type of scheduler that disco should use. The only options are fair and fifo. Default is fair.

DISCO_SCHEDULER_ALPHA

Parameter controlling how much the fair scheduler punishes long-running jobs vs. short ones. Default is .001 and should usually not need to be changed.

Settings used by the testing environment:

DISCO_TEST_DISCODB

Whether or not to run discodb tests. Default is ''.

DISCO_TEST_HOST

The hostname that the test data server should bind on. Default is DISCO_MASTER_HOST.

DISCO_TEST_PORT

The port that the test data server should bind to. Default is 9444.

Settings used by DDFS:

DDFS_ROOT

Deprecated since version 0.4.

Use DDFS_DATA instead. Only provided as a default for backwards compatability. Default is obtained using os.path.join(DISCO_ROOT, 'ddfs').

DDFS_DATA

The root data directory for DDFS. Default is obtained using DDFS_ROOT.

DDFS_PUT_PORT

The port to use for writing to DDFS nodes. Must be open to the Disco client unless proxying is used. Default is 8990.

DDFS_PUT_MAX

The maximum default number of retries for a PUT operation. Default is 3.

DDFS_GET_MAX

The maximum default number of retries for a GET operation. Default is 3.

DDFS_READ_TOKEN

The default read authorization token to use. Default is None.

DDFS_WRITE_TOKEN

The default write authorization token to use. Default is None.

DDFS_GC_INITIAL_WAIT

The amount of time to wait after startup before running GC (in minutes). Default is '', which triggers an internal default of 5 minutes.

DDFS_GC_BALANCE_THRESHOLD
The distance a node's disk utilization can be from the average
disk utilization of the cluster before the node is considered
to be over-utilized or under-utilized. Default is ``0.1``.
DDFS_PARANOID_DELETE

Instead of deleting unneeded files, DDFS garbage collector prefixes obsolete files with !trash., so they can be safely verified/deleted by an external process. For instance, the following command can be used to finally delete the files (assuming that DDFS_DATA = "/srv/disco/ddfs"):

find /srv/disco/ddfs/ -perm 600 -iname '!trash*' -exec rm {} \;

Default is ''.

The following settings are used by DDFS to determine the number of replicas for data/metadata to keep (it is not recommended to use the provided defaults in a multinode cluster):

DDFS_TAG_MIN_REPLICAS

The minimum number of replicas for a tag operation to succeed. Default is 1.

DDFS_TAG_REPLICAS

The number of replicas of tags that DDFS should aspire to keep. Default is 1.

DDFS_BLOB_REPLICAS

The number of replicas of blobs that DDFS should aspire to keep. Default is 1.

DDFS_SPACE_AWARE

Whether DDFS should take the amount of free space in the nodes into account when choosing the nodes to write to. Default is ````.

DDFS_ABSOLUTE_SPACE

Only effective in the space-aware mode. If set, the nodes with the higher absolute free space will be given precedence for hosting replicas. If unset, the nodes with the highest ratio of the free space to the total space will be given precedence for hosting the replicas.