Warning

This document is for an in-development version of Galaxy. You can alternatively view this page in the latest release if it exists or view the top of the latest release's documentation.

Scaling and Load Balancing

The Galaxy framework is written in Python and makes extensive use of threads. However, one of the drawbacks of Python is the Global Interpreter Lock, which prevents more than one thread from being on CPU at a time. Because of this, having a multi-core system will not improve the Galaxy framework’s performance out of the box since Galaxy can use (at most) one core at a time in its default configuration. However, Galaxy can easily run in multiple separate processes, which solves this problem. For a more thorough explanation of this problem and why you will almost surely want to switch to the multiprocess configuration if running for more than a small handful of users, see the production configuration page.

Just to be clear: increasing the values of threadpool_workers in galaxy.yml or the number of plugin workers in job_conf.xml will not make you Galaxy server much more responsive. The key to scaling Galaxy is the ability to run multiple Galaxy servers which co-operatively work on the same database.

Terminology

  • web worker - Galaxy server process responsible for servicing web requests for the UI/API
  • job handler - Galaxy server process responsible for setting up, starting, and monitoring jobs, submitting jobs to a cluster (if configured), for setting metadata (if not set on the cluster), and cleaning up after jobs
    • tags - Handlers can be grouped in to a “pool” of handlers using tags, after which, individual tools may be mapped to a handler tag such that all executions of that tool are handled by the tagged handler(s).
    • default - Any handlers without defined tags - aka “untagged handlers” - will handle executions of all tools not mapped to a specific handler ID or tag.
  • uWSGI - Powerful application server written in C that implements the HTTP and Python WSGI protocols
    • Mules - uWSGI processes started after the main application (Galaxy) that can run separate code and receive messages from uWSGI web workers
    • Zerg Mode - uWSGI configuration where multiple copies of the same application can be started simultaneously in order to maintain availability during application restarts
  • Webless Galaxy application - The Galaxy application run as a standalone Python application with no web/WSGI server

Application Servers

It is possible to run the Galaxy server in many different ways, including under different web application frameworks, or as a standalone server with no web stack. For most of its modern life, prior to the 18.01 release, Galaxy (by default) used the Python Paste web stack, and ran in a single process.

Beginning with Galaxy release 18.01, the default application server for new installations of Galaxy is uWSGI. Prior to 18.01, it was possible (and indeed, recommended for production Galaxy servers) to run Galaxy under uWSGI, but it was necessary to install and configure uWSGI separately from Galaxy. uWSGI is now provided with Galaxy as a Python Wheel and installed in to its virtualenv, as described in detail in the Framework Dependencies documentation.

uWSGI has numerous benefits over Python Paste for our purposes:

  • Written in C and designed to be high performance
  • Easily runs multiple processes by increasing processes config option
  • Load balances multiple processes internally rather than requiring load balancing in the proxy server
  • Offload engine for serving static content
  • Speaks high performance native protocol between uWSGI and proxy server
  • Can speak HTTP and HTTPS protocols without proxy server
  • Incredibly featureful, supports a wide array of deployment scenarios
  • Supports WebSockets, which enable Galaxy Interactive Environments out-of-the-box without a proxy server or Node.js

Deployment Options

There are multiple deployment strategies for the Galaxy application that you can choose from. The right one depends on the configuration of the infrastructure on which you are deploying. In all cases, all Galaxy job features such as running on a cluster are supported.

Although uWSGI implements nearly all the features that were previously the responsibility of an upstream proxy server, at this time, it is still recomended to place a proxy server in front of uWSGI and utilize it for all of its traditional roles (serving static content, serving dataset downloads, etc.) as described in the production configuration documentation.

When using uWSGI with a proxy server, it is recommended that you use the native high performance uWSGI protocol (supported by both Apache and nginx) between uWSGI and the proxy server, rather than HTTP.

uWSGI with jobs handled by web workers (default configuration)

Referred to in this documentation as the uWSGI all-in-one strategy.

  • Job handlers and web workers are the same processes and cannot be separated
  • The web worker that receives the job request from the UI/API will be the job handler for that job

Under this strategy, jobs will be handled by uWSGI web workers. Having web processes handle jobs will negatively impact UI/API performance.

This is the default out-of-the-box configuration as of Galaxy Release 18.01.

uWSGI for web serving with Mules as job handlers

Referred to in this documentation as the uWSGI + Mules strategy.

  • Job handlers run as children of the uWSGI process
  • Jobs are dispatched from web workers to job handlers via native mule messaging
  • Jobs can only be dispatched to mules on the same host
  • Trivially easy to enable (disabled by default for simplicity reasons)

Under this strategy, job handling is offloaded to dedicated non-web-serving processes that are started and stopped directly by the master uWSGI process. As a benefit of using mule messaging, only job handlers that are alive will be selected to run jobs.

This is the recommended deployment strategy.

Important

If using Zerg Mode or running more than one uWSGI master process, do not use uWSGI + Mules. Doing so can can cause jobs to be executed by mutiple handlers when recovering unassigned jobs at Galaxy server startup.

Multiple master processes is a rare configuration and is typically only used in the case of load balancing the web application across multiple hosts. Note that multiple master proceses is not the same thing as the processess uWSGI configuration option, which is perfectly safe to set when using job handler mules.

For these scenarios, uWSGI + Webless is the recommended deployment strategy.

uWSGI for web serving and Webless Galaxy applications as job handlers

Referred to in this documentation as the uWSGI + Webless strategy.

  • Job handlers are started as standalone Python applications with no web stack
  • Jobs are dispatched from web workers to job handlers via the Galaxy database
  • Jobs can be dispatched to job handlers running on any host
  • Additional job handlers can be added dynamically without reconfiguring/restarting Galaxy (19.01 or later)
  • The recommended deployment strategy for production Galaxy instances prior to 18.01

Like mules, under this strategy, job handling is offloaded to dedicated non-web-serving processes, but those processes are managed by the administrator.

By default, a handler is randomly assigned by the web worker when the job is submitted via the UI/API, meaning that jobs may be assigned to dead handlers. However, beginning in Galaxy release 19.01, new job handler assignment methods are available that allow handlers to self-assign jobs (a handler is no longer required to be assigned when the job is created).

This is the recommended deployment strategy when Zerg Mode is used, for Galaxy servers that run web servers and job handlers on different hosts, and for deployments where dynamic handler addition is desired.

Beginning with Galaxy release 19.01, it is also possible to use a combination of both uWSGI + Mules and uWSGI + Webless, referred to in the documentation as uWSGI + Hybrid.

Legacy

Other deployment strategies were commonly used in older versions of Galaxy and can be found in previous versions of the documentation, but these are deprecated and should no longer be used.

Job Handler Assignment Methods

Prior to Galaxy release 19.01, the method by which handlers were selected to be assigned to jobs was dependent on your job configuration and use of uWSGI Mules, and was not configurable by the administrator. Beginning with Galaxy release 19.01, two new handler assignment methods have been added and methods are now configurable with the assign_with attribute on the <handlers> tag in job_conf.xml. The available methods are:

  • Database Self Assignment - Like In-memory Self Assignment but assignment occurs by setting a new job’s ‘handler’ column in the database to the process that created the job at the time it is created. Additionally, if a tool is configured to use a specific handler (ID or tag), that handler is assigned (tags by Database Preassignment). This is the default if no handlers are defined and no job-handlers uWSGI Farm is present (the default for a completely unconfigured Galaxy).
  • In-memory Self Assignment - Jobs are assigned to the web worker that received the tool execution request from the user via an internal in-memory queue. If a tool is configured to use a specific handler, that configuration is ignored; the process that creates the job always handles it. This can be slightly faster than Database Self Assignment but only makes sense in single process environments without dedicated job handlers. This option supercedes the former track_jobs_in_database option in galaxy.yml and corresponds to setting that option to false.
  • Database Preassignment - Jobs are assigned a handler by selecting one at random from the configured tag or default handlers at the time the job is created. This occurs by the web worker that receives the tool execution request (via the UI or API) setting a new job’s ‘handler’ column in the database to the randomly chose handler ID (hence “preassignment”). This is the default if handlers are defined and no job-handlers uWSGI Farm is present.
  • uWSGI Mule Messaging - Jobs are assigned a handler via uWSGI mule messaging. A mule in the job-handlers (for default/untagged tool-to-handler mappings) or job-handlers.<tag> farm will receive the message and assign itself. This the default if a job-handlers uWSGI Farm is present and no handlers are configured.
  • Database Transaction Isolation (new in 19.01) - Jobs are assigned a handler by handlers selecting the unassigned job from the database using SQL transaction isolation, which uses database locks to guarantee that only one handler can select a given job. This occurs by the web worker that receives the tool execution request (via the UI or API) setting a new job’s ‘handler’ column in the database to the configured tag/default (or _default_ if no tag/default is configured). Handlers “listen” for jobs by selecting jobs from the database that match the handler tag(s) for which they are configured.
  • Database SKIP LOCKED (new in 19.01) - Jobs are assigned a handler by handlers selecting the unassigned job from the database using SELECT ... FOR UPDATE SKIP LOCKED on databases that support this query (see the next section for details). This occurs via the same process as Database Transaction Isolation, the only difference is the way in which handlers query the database.

In the event that both a job-handlers uWSGI Farm is present and handlers are configured, the default is uWSGI Mule Messaging followed by Database Preassignment. At present, only uWSGI Mule Messaging is capable of deferring handler assignment to a later method (which would occur in the event that a tool is configured to use a tag for which there is not a matching farm).

In all cases, if a tool is configured to use a specific handler (by ID, not tag), configured assignment methods are ignored and that handler is directly assigned in the job’s ‘handler’ column at job creation time.

See the config/job_conf.xml.sample_advanced file in the Galaxy distribution for instructions on setting the assignment method.

Choosing an Assignment Method

Prior to Galaxy 19.01, the most common deployment strategies (e.g. uWSGI + Webless) assigned handlers using what is now (since 19.01) referred to as Database Preassignment. Although still the default in many cases (until the new methods mature), preassignment has a few drawbacks:

  • Web workers do not have a way to know whether a particular handler is alive when assigning that handler
  • Jobs are not load balanced across handlers
  • Changing the number of handlers requires changing job_conf.xml and restarting all Galaxy processes

The new “database locking” methods (Database SKIP LOCKED and Database Transaction Isolation) were created to solve these issues. The preferred method between the two new options is Database SKIP LOCKED, but it requires PostgreSQL 9.5 or newer, MySQL 8.0 or newer (untested), or MariaDB 10.3 or newer (untested). If using an older database version, use Database Transaction Isolation instead. A detailed explanation of these database locking methods in PostgreSQL can be found in the excellent What is SKIP LOCKED for in PostgreSQL 9.5? entry on the 2ndQuadrant PostgreSQL Blog.

The preferred method depends on your deployment strategy:

  • uWSGI + Mules - uWSGI Mule Messaging is preferred.
  • uWSGI + Webless - Either Database SKIP LOCKED or Database Transaction Isolation is preferred.
  • uWSGI + Hybrid - Either Database SKIP LOCKED or Database Transaction Isolation is preferred. If your mule and webless handlers are in non-overlapping pools (i.e. tags, or untagged), you can alternatively use both uWSGI Mule Messaging followed by either Database SKIP LOCKED or Database Transaction Isolation. If pools overlap, using uWSGI Mule Messaging would prevent any non-mule handlers in that pool from being assigned jobs.

Handlers (as well as assignment methods) are not configurable when using uWSGI all-in-one.

Configuration

uWSGI

Although this document goes in to significant detail about uWSGI configuration, many more options are available, as well as additional documentation about options described here. Consult the uWSGI documentation for more:

Configuration is performed in the uwsgi section of galaxy.yml. You will find that the default, if copied from galaxy.yml.sample, is commented out. The default configuration options are provided to uWSGI on the command line by Galaxy’s run.sh script.

Galaxy releases prior to 18.01 (or upgraded-to-18.01+ servers which have not migrated their configuration to the YAML format) used an INI-format configuration file, galaxy.ini.

Note that uWSGI’s YAML parser is hand-coded and not actually conformant to the YAML standard. Specifically:

  • Multiple identical keys with unique values can exist in the same dictionary/hash, as with hook-master-start in the example below.
  • Quoting values (with single or double quotes) is unncessary since the parser treats all values as strings. The parser does not correctly handle these quote characters, resulting in invalid values.

If using galaxy.ini, the option names and values are the same but in INI format, for example:

[uwsgi]
processes = 4
socket = 127.0.0.1:4001
...

Configuration common to all uWSGI deployment styles

In galaxy.yml, define a uwsgi section. Shown below are the options common to all deployment strategies:

uwsgi:

  # required in order to start the galaxy application
  module: galaxy.webapps.galaxy.buildapp:uwsgi_app()
  virtualenv: .venv
  pythonpath: lib

  # performance options
  master: true
  enable-threads: true
  processes: 2
  threads: 4
  offload-threads: 1

  # fix up signal handling
  die-on-term: true
  hook-master-start: unix_signal:2 gracefully_kill_them_all
  hook-master-start: unix_signal:15 gracefully_kill_them_all

  # listening options

  # job handling options

Some of these options warrant explanation:

  • master: Instructs uWSGI to first start a master process manager and then fork web workers, mules, http servers (if enabled), and any others from the master. This is required for certain operational modes such as daemonization, but can interfere with the use of <CTRL>+<C> to shut down Galaxy when running in the foreground on the command line, and so is not enabled by default (except when run.sh --daemon is used). Its use is strongly recommended for all production deployments.
  • processes: Controls the number of Galaxy application processes uWSGI will spawn. Increased web performance can be attained by increasing this value.
  • threads: Controls the number of web worker threads each application process will spawn.
  • offload-threads: uWSGI can use a dedicated threadpool for serving static content and handling internal routing, setting this value automatically enables such offloading.

Additional options are explained in the uWSGI Minutiae below.

Note that the performance option values given above are just examples and should be tuned per your specific needs. However, as given, they are a good place to start.

Due to the Python GIL, increasing the value of threads has diminishing returns on web performance while increasing the memory footprint of each application process. Increasing it is most useful on servers experiencing a high amount of IO waiting, but the greatest performance gain comes from increasing processes as appropriate for the hardware on which Galaxy is running.

Listening and proxy options

With a proxy server:

To use the native uWSGI protocol, set the socket option:

  # listening options
  socket: /srv/galaxy/var/uwsgi.sock

Here we’ve used a UNIX domain socket because there’s less overhead than a TCP socket and it can be secured by filesystem permissions, but you can also listen on a port:

  # listening options
  socket: 127.0.0.1:4001

The choice of port 4001 is arbitrary, but in both cases, the socket location must match whatever socket the proxy server is configured to communicate with. If using a UNIX domain socket, be sure that the proxy server’s user has read/write permission on the socket. Because Galaxy and the proxy server most likely run as different users, this is not likely to be the case by default. One common solution is to add the proxy server’s user to the Galaxy user’s primary group. uWSGI’s chmod-socket option can also help here.

You can consult the Galaxy documentation for Apache or nginx for help with the proxy-side configuration.

By setting the socket option, run.sh will no longer automatically serve Galaxy via HTTP (since it is assumed that you are setting a socket to serve Galaxy via a proxy server). If you wish to continue serving HTTP directly with uWSGI while socket is set, you can use the http option as shown in the directions below.

Without a proxy server or with a proxy server that does not speak the uWSGI native protocol:

uWSGI can be configured to serve HTTP and/or HTTPS directly:

  # listening options
  http: :8080
  https: :8443,server.crt,server.key
  static-map: /static/style=static/style/blue
  static-map: /static=static

To bind to ports < 1024 (e.g. if you want to bind to the standard HTTP/HTTPS ports 80/443), you must bind as the root user and drop privileges to the Galaxy user with a configuration such as:

  # listening options
  shared-socket: :80
  shared-socket: :443,server.crt,server.key
  http: =0
  https: =1
  uid: galaxy
  gid: galaxy
  static-map: /static/style=static/style/blue
  static-map: /static=static

To redirect HTTP traffic to the HTTPS port rather than serving Galaxy over HTTP, change http: =0 in the above example to http-to-https: =0.

Because run.sh performs setup steps, it should not be run as root. Instead, you can run uWSGI directly as root with:

# cd /srv/galaxy/server
# ./.venv/bin/uwsgi --yaml config/galaxy.yml

You can run the startup-time setup steps as the galaxy user after upgrading Galaxy with sh ./scripts/common_startup.sh.

Job Handling

Warning

In all strategies, once a handler has been assigned jobs, you cannot unconfigure that handler (e.g. to decrease the number of handlers) until it has finished processing all its assigned jobs, or else its jobs will never reach a terminal state. In order to allow a handler to run but not receive any new jobs, configure it with an unused tag (e.g <handler id="handler5" tags="drain" />) and restart all Galaxy processes.

Alternatively, you can stop the handler and reassign its jobs to another handler, but this is currently only possible using an UPDATE query in the database and is only recommended for advanced Galaxy administrators.

uWSGI all-in-one job handling

Ensure that no <handlers> section exists in your job_conf.xml (or no job_conf.xml exists at all) and start Galaxy normally. No additional configuration is required. To increase the number of web workers/job handlers, increase the value of processes. Jobs will be handled by the web worker that receives the job setup request (via the UI/API).

Note

If a <handlers> section is defined in job_conf.xml, Galaxy’s web workers will no longer load and start the job handling code, so tools cannot be mapped to specific handlers in this strategy. If you wish to control job handling, choose another deployment strategy.

uWSGI + Mule job handling

Ensure that no <handlers> section exists in your job_conf.xml (or no job_conf.xml exists at all) and add the following to the uwsgi section of galaxy.yml to start a single job handler mule:

  # job handling options
  mule: lib/galaxy/main.py
  farm: job-handlers:1

Then start Galaxy normally. To add additional mule handlers, add additional mule options and add their ID(s), comma separated, to the job-handlers farm. For example, 3 handlers are defined like so:

  # job handling options
  mule: lib/galaxy/main.py
  mule: lib/galaxy/main.py
  mule: lib/galaxy/main.py
  farm: job-handlers:1,2,3

By default, a job will be handled by whatever mule currently has the lock on the mule message queue. After receiving a message, it will release the lock, giving other mules a chance to handle future jobs. Jobs can be explicitly mapped to specific mules as described in the Job configuration documentation by using the handler IDs main.job-handlers.N, where N is the mule’s position in the farm, starting at 1 and incrementing for each mule in the farm (this is not necessarily the mule ID, but it will be if you only define one farm and you add mules to that farm in sequential order). Each worker that you wish to explicitly map jobs to should be defined in the <handlers> section of job_conf.xml. Do not define a default handler.

For example, to have the 2nd mule in the three-mule job-handlers farm shown above handle the test1 tool, you would set the following in job_conf.xml (irrelevant options are not shown):

<job_conf>
    <handlers>
        <handler id="main.job-handlers.2" />
    </handlers>
    <tools>
        <tool id="test1" handler="main.job-handlers.2" />
    </tools>
</job_conf>

It is also possible to create separate pools for handler tags: the farm name for tags is simply job-handlers.<tag>. In the following example, all jobs will be handled by mules 1 and 2, except for executions of tool test1: those jobs will be handled by mule 3.

The uwsgi section in galaxy.yml:

  # job handling options
  mule: lib/galaxy/main.py
  mule: lib/galaxy/main.py
  mule: lib/galaxy/main.py
  farm: job-handlers:1,2
  farm: job-handlers.special:3

In job_conf.xml:

<job_conf>
    <tools>
        <tool id="test1" handler="special" />
    </tools>
</job_conf>

uWSGI + Webless job handling

Beginning with Galaxy release 19.01, webless handlers do not always need to be defined in job_conf.xml.

Statically defined handlers

In the <handlers> section in job_conf.xml, define the webless handlers you plan to start. Tools can be mapped to specific handlers, or to handler tags, as in the following example:

<job_conf>
    <handlers>
        <handler id="handler1" />
        <handler id="handler2" />
        <handler id="handler3" tags="nodefault" />
        <handler id="handler4" tags="special" />
        <handler id="handler5" tags="special" />
    </handlers>
    <tools>
        <tool id="test1" handler="handler3" />
        <tool id="test2" handler="special" />
        <tool id="test3" handler="handler2" />
    </tools>
</job_conf>

Tip

Any untagged handler will be automatically considered a default handler. As seen in the example above, it is possible to map any tool to any handler or tag, however, a handler must be tagged to prevent it from handling jobs created for tools that are not explicitly mapped to handlers. Thus, handler2 will handle all executions of tool test3, but it will also (along with handler1) handle tools that are not explicitly mapped to handlers. In contrast, handler3 will only handle executions of tool test1.

run.sh will start the uWSGI process(es), but you will need to start the webless handler processes yourself. This is done on the command line like so:

$ cd /srv/galaxy/server
$ ./scripts/galaxy-main -c config/galaxy.yml --server-name handler0 --daemonize
$ ./scripts/galaxy-main -c config/galaxy.yml --server-name handler1 --daemonize
$ ./scripts/galaxy-main -c config/galaxy.yml --server-name handler2 --daemonize

However, a better option to managing processes by hand is to use a process manager as documented in the Starting and Stopping section.

Dynamically defined handlers

In order to define handlers dynamically, you must be using one of the new “database locking” handler assignment methods as explained in [Job Handler Assignment Methods][#job-handler-assignment-methods], such as in the following job_conf.xml:

<job_conf>
    <handlers assign_with="db-skip-locked" />
    <tools>
        <tool id="test1" handler="special" />
    </tools>
</job_conf>

As with statically defined handlers, run.sh will start the uWSGI process(es), but you will need to start the webless handler processes yourself. This is done on the command line like so (note the addition of the --attach-to-pool option):

$ cd /srv/galaxy/server
$ ./scripts/galaxy-main -c config/galaxy.yml --server-name handler0 --attach-to-pool job-handlers --daemonize
$ ./scripts/galaxy-main -c config/galaxy.yml --server-name handler1 --attach-to-pool job-handlers.special --daemonize
$ ./scripts/galaxy-main -c config/galaxy.yml --server-name handler2 --attach-to-pool job-handlers --attach-to-pool job-handlers.special --daemonize

In this example:

  • handler0 and handler2 will handle tool executions that are not explicitly mapped to handlers
  • handler1 and handler2 will handle tool executions that are mapped to the special handler tag

uWSGI + Hybrid job handling

Follow the process for both uWSGI + Mules and uWSGI + Webless:

  1. Define Mules as in uWSGI + Mules
  2. Define Webless handlers in <handlers> as in uWSGI + Webless (when using static handlers)
  3. Configure Webless handlers to start as in uWSGI + Webless

uWSGI Minutiae

Threads

Although enable-threads was explicitly set in our example, in reality, as long as any value is set for the threads option, enable-threads is set implicitly. This option enables the Python GIL and application threads (threads started by Galaxy itself for various non-web tasks), which Galaxy uses extensively. Setting it explicitly, however, is harmless and can prevent strange difficult-to-debug situations if threads is accidentally unset.

Worker/Mule shutdown/reload mercy

By default, uWSGI will wait up to 60 seconds for web workers and mules to terminate. This is generally safe for servicing web requests, but some parts of Galaxy’s job preparation/submission and collection/finishing operations can take quite a bit of time to complete and are not entirely reentrant: job errors or state inconsistencies can occur if interrupted (although every effort has been made to minimize such possibilities). By default, Galaxy will wait up to 30 seconds for the threads allocated for these operations to terminate after instructing them to shut down. You can change this behavior by increasing the value of monitor_thread_join_timeout in the galaxy section of galaxy.yml. The maximum amount of time that Galaxy will take to shut down job runner workers is monitor_thread_join_timeout * runner_plugin_count since each plugin is shut down sequentially (runner_plugin_count is the number of <plugin>s in your job_conf.xml).

Thus you should set the appropriate uWSGI *-restart-mercy option to a value higher than the maximum job runner worker shutdown time. If using uWSGI all-in-one, set worker-reload-mercy, and if using uWSGI + Mule job handling, set mule-reload-mercy (both in the uwsgi section of galaxy.yml).

Signals

The signal handling options (die-on-term and hook-master-start with unix_signal values) are not required but, if set, will override uWSGI’s unconventional signal handling and cause SIGTERM to kill the server rather than restart it, and the uWSGI master process to gracefully shut down its web workers and job handler mules (i.e. the various Galaxy application processes) when it receives a SIGINT (signal 2) or SIGTERM (signal 15) signal (e.g. from kill(1) or <CTRL>+<C>). When shutting down gracefully, uWSGI will wait 60 seconds (by default, but this can be changed with the reload-mercy, worker-reload-mercy and mule-reload-mercy options) for child processes to die before forcefully killing them with SIGKILL (signal 9). Alternatively, you may prefer to have it shut down gracefully on SIGTERM but forcefully on SIGINT (forceful shutdown by uWSGI is still slightly cleaner than kill -9 of the master process since it can attempt to release sockets cleanly) or vice-versa, which you can do by setting one of the signals to call kill_them_all rather than gracefully_kill_them_all:

  # fix up signal handling
  die-on-term: true
  hook-master-start: unix_signal:2 kill_them_all
  hook-master-start: unix_signal:15 gracefully_kill_them_all

More details on the unix_signal hook can be found in uWSGI Issue #849.

Logging and daemonization

It’s possible to configure uWSGI to log to a file with the logto or logto2 options (when running in the foreground, the default), but more advanced logging options that split log files for each process are possible and described in the Galaxy Logging Configuration documentation

When running as a daemon with run.sh --daemon, output is logged to galaxy.log and the pid is written to galaxy.pid. These can be controlled with the daemonize and pidfile arguments (their daemonize2 and pidfile2 counterparts wait until after the application successfully loads to open and write the files). If you set a daemonize* option, you should not use the --daemon argument to run.sh (or not use run.sh at all, and start Galaxy directly with uwsgi on the command line).

External uwsgi binary:

It is still possible to run Galaxy using an external copy of uWSGI (for example, installed from APT under Debian/Ubuntu). This was the recommended installation method in the past. To use an external uWSGI, you’ll need simply need to start with uWSGI directly, rather than using the run.sh script. Once you have configured Galaxy/uWSGI, you can start it with:

$ cd /srv/galaxy/server
$ uwsgi --yaml config/galaxy.yml

When installing uWSGI, be sure to install the Python plugin, as this is not always contained in the same package (such as when installing from APT under Debian/Ubuntu). With the APT packages, you will also need to add --plugin python to the command line (or plugin: python to the uwsgi section of galaxy.yml).

Other options

The py-call-osafterfork option may only be needed when mule messaging is in use, but it seems to have no negative effect and may solve other situations with starting a complex threaded Python application.

The chdir option is useful if you are not using run.sh, to be able to call uwsgi from anywhere without having to cd to the Galaxy directory first.

Monitoring

The uwsgitop tool uses uWSGI’s stats server (the stats option, which is configured to listen on a socket or port in the same manner as the socket option) to report on the health and performance of the web workers. It can be installed with pip install uwsgitop. This tool can be useful for determing whether a worker is stuck, or seeing the throughput of traffic on your site.

Starting and Stopping

If you are using the uWSGI + Webless job handlers deployment strategy or want to run your Galaxy server as a persistent service, you can control it through a process manager. The current recommended process manager is Supervisord. If you are comfortable with systemd and are running a relatively modern Linux distribution, you can also configure Galaxy as a service directly in systemd.

Supervisord

You can use a supervisord config file like the following or be inspired by this example. Be sure to supervisord update or supervisord reread && supervisord restart whenever you make configuration changes.

[program:web]
command         = /srv/galaxy/venv/bin/uwsgi --yaml /srv/galaxy/config/galaxy.yml
directory       = /srv/galaxy/server
umask           = 022
autostart       = true
autorestart     = true
startsecs       = 10
stopwaitsecs    = 65
user            = galaxy
numprocs        = 1
stopsignal      = INT

This configuration defines a “program” named web which represents our Galaxy uWSGI frontend. You’ll notice that we’ve set a command, a directory, a umask, all of which you should be familiar with. Additionally we’ve specified that the process should autostart on boot, and autorestart if it ever crashes. We specify startsecs to say “the process must stay up for this long before we consider it OK. If the process crashes sooner than that (e.g. bad changes you’ve made to your local installation) supervisord will try again a couple of times to restart the process before giving up and marking it as failed. This is one of the many ways supervisord is much friendly for managing these sorts of tasks.

The value of stopwaitsecs should be at least as large as the smallest value of uWSGI’s reload-mercy, worker-reload-mercy, and mule-reload-mercy options, all of which default to 60.

If using the uWSGI + Webless strategy, you’ll need to addtionally define job handlers to start. There’s no simple way to activate a virtualenv when using supervisor, but you can simulate the effects by setting $PATH and $VIRTUAL_ENV:

[program:handler]
command         = /srv/galaxy/venv/bin/python ./scripts/galaxy-main -c /srv/galaxy/config/galaxy.yml --server-name=handler%(process_num)s --pid-file=/srv/galaxy/var/handler%(process_num)s.pid --log-file=/srv/galaxy/log/handler%(process_num)s.log
directory       = /srv/galaxy/server
process_name    = handler%(process_num)s
numprocs        = 3
umask           = 022
autostart       = true
autorestart     = true
startsecs       = 15
stopwaitsecs    = 35
user            = galaxy
environment     = VIRTUAL_ENV="/srv/galaxy/venv",PATH="/srv/galaxy/venv/bin:%(ENV_PATH)s"

This is similar to the “web” definition above, however, you’ll notice that we use %(process_num)s. That’s a variable substitution in the command and process_name fields. We’ve set numprocs = 3, which says to launch three handler processes. Supervisord will loop over 0..numprocs and launch handler0, handler1, and handler2 processes automatically for us, templating out the command string so each handler receives a different log file and name.

The value of stopwaitsecs should be at least as large as monitor_thread_join_timeout * runner_plugin_count, which is 30 in the default configuration (monitor_thread_join_timeout is a Galaxy configuration option and runner_plugin_count is the number of <plugin>s in your job_conf.xml).

Lastly, collect the tasks defined above into a single group. If you are not using webless handlers this is as simple as:

[group:galaxy]
programs = web

With webless handlers, it is:

[group:galaxy]
programs = web, handler

This will let us manage these tasks more globally with the supervisorctl command line tool:

# supervisorctl status
galaxy:handler0                  RUNNING   pid 7275, uptime 16:32:17
galaxy:handler1                  RUNNING   pid 7276, uptime 16:32:17
galaxy:handler2                  RUNNING   pid 7277, uptime 16:32:17
galaxy:web                       RUNNING   pid 7299, uptime 16:32:16

This command shows us the status of our jobs, and we can easily restart all of the processes at once by naming the group. Familiar commands like start and stop are also available.

# supervisorctl restart galaxy:
galaxy:handler0: stopped
galaxy:handler1: stopped
galaxy:handler2: stopped
galaxy:web: stopped
galaxy:web: started
galaxy:handler0: started
galaxy:handler1: started
galaxy:handler2: started

Systemd

TODO: write this section.

Transparent Restart - Zerg Mode

The standard uWSGI operation mode allows you to restart the Galaxy application while blocking client connections. Zerg Mode does away with the waiting by running a special Zerg Pool process, and connecting Zergling workers (aka Galaxy application processes) to the pool. As long as at least one is connected, requests can be served.

See the GCC2017 Admin Training session on how to set this up.