Zero Downtime Deploys
New
Introduced in 0.5.0
checks <app> Show zero-downtime status
checks:disable <app> Disable zero-downtime checks
checks:enable <app> Enable zero-downtime checks
Following a deploy, dokku will wait 10 seconds before routing traffic to the new container to give your application time to boot up. If the application is not running after this time, then the deploy is failed and your old container will continue serving traffic. You can modify this value globally or on a per-application basis:
dokku config:set --global DOKKU_DEFAULT_CHECKS_WAIT=30
dokku config:set <app> DOKKU_DEFAULT_CHECKS_WAIT=30
You can also choose to skip checks completely on a per-application basis:
Dokku will wait 60 seconds before stopping the old container so that existing connections are given a chance to complete. You can modify this value globally or on a per-application basis:
Info
Note that during this time, multiple containers may be running on your server, which can be an issue for memory-hungry applications on memory-constrained servers.
Checks
If your application needs a longer period to boot up - perhaps to load data into memory, or because of slow boot time - you may also use dokku's checks functionality to more precisely check whether an application can serve traffic or not.
Checks are run against the detected web process from your application's Procfile. For non-web processes, dokku will fallback to the aforementioned process uptime check.
To specify checks, add a CHECKS file to the root of your project directory. The CHECKS file should be plain text and may contain:
- Check instructions
- Settings (NAME=VALUE)
- Comments (lines starting with #)
- Empty lines
Info
For dockerfile-based deploys, the file must be in /app/CHECKS within the container. /app is used by default as the root container directory for buildpack-based deploys.
Check Instructions
The format of a check instruction is a path or relative URL, optionally followed by the expected content:
The CHECKS file can contain multiple checks:
To check an application that supports multiple hostnames, use relative URLs that include the hostname:
You can also specify the protocol to explicitly check HTTPS requests:
Check Settings
The default behavior is to wait for 5 seconds before running the checks, to timeout the checks after 30 seconds, and to attempt the checks 5 times. If the checks fail 5 times, the deployment is considered failed and the old container will continue serving traffic.
You can change the default behavior by setting WAIT, TIMEOUT, and ATTEMPTS to different values in the CHECKS file:
WAIT=30 # Wait 1/2 minute
TIMEOUT=60 # Timeout after a minute
ATTEMPTS=10 # Attempt checks 10 times
/ My Amazing App
You can also override the default WAIT, TIMEOUT, and ATTEMPTS variables for the global dokku installation:
dokku config:set --global DOKKU_CHECKS_WAIT=30
dokku config:set --global DOKKU_CHECKS_TIMEOUT=60
dokku config:set --global DOKKU_CHECKS_ATTEMPTS=10
If your application runs multiple processes (a background worker configured in your Procfile, for example) and you have checks to ensure that your web application has booted up, you may want to disable the default check wait time for that application to avoid the 10 second wait per non-web process:
Example: Successful Rails Deployment
In this example, a Rails application is successfully deployed to dokku. The initial round of checks fails while the server is starting, but once it starts they succeed and the deployment is successful. WAIT is set to 10 because our application takes a while to boot up. ATTEMPTS is set to 6, but the third attempt succeeds.
CHECKS file
For this check to work, we've added a line to config/routes.rb that simply returns a string:
Deploy Output
Note
The output has been trimmed for brevity
git push dokku master
-----> Cleaning up...
-----> Building myapp from herokuish...
-----> Adding BUILD_ENV to build environment...
-----> Ruby app detected
-----> Compiling Ruby/Rails
-----> Using Ruby version: ruby-2.0.0
.....
-----> Discovering process types
Procfile declares types -> web
-----> Releasing myapp...
-----> Deploying myapp...
-----> Running pre-flight checks
-----> Attempt 1/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/check.txt => "simple_check"
!
curl: (7) Failed to connect to 172.17.0.155 port 5000: Connection refused
! Check attempt 1/6 failed.
-----> Attempt 2/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/check.txt => "simple_check"
!
curl: (7) Failed to connect to 172.17.0.155 port 5000: Connection refused
! Check attempt 2/6 failed.
-----> Attempt 3/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/check.txt => "simple_check"
-----> All checks successful!
=====> myapp container output:
=> Booting Thin
=> Rails 4.2.0 application starting in production on http://0.0.0.0:5000
=> Run `rails server -h` for more startup options
=> Ctrl-C to shutdown server
Thin web server (v1.6.3 codename Protein Powder)
Maximum connections set to 1024
Listening on 0.0.0.0:5000, CTRL+C to stop
=====> end myapp container output
-----> Running post-deploy
-----> Configuring myapp.dokku.example.com...
-----> Creating http nginx.conf
-----> Running nginx-pre-reload
Reloading nginx
-----> Shutting down old container in 60 seconds
=====> Application deployed:
http://myapp.dokku.example.com
Example: Failing Rails Deployment
In this example, a Rails application fails to deploy. The reason for the failure is that the postgres database connection fails. The initial checks will fail while we wait for the server to start up, just like in the above example. However, once the server does start accepting connections, we will see an error 500 due to the postgres database connection failure.
Once the attempts have been exceeded, the deployment fails and we see the container output, which shows the Postgres connection errors.
CHECKS file
Info
The check to the root url '/' would normally access the database.
Deploy Output
Note
The output has been trimmed for brevity
git push dokku master
-----> Cleaning up...
-----> Building myapp from herokuish...
-----> Adding BUILD_ENV to build environment...
-----> Ruby app detected
-----> Compiling Ruby/Rails
-----> Using Ruby version: ruby-2.0.0
.....
Discovering process types
Procfile declares types -> web
Releasing myapp...
Deploying myapp...
Running pre-flight checks
-----> Attempt 1/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/ => ""
!
curl: (7) Failed to connect to 172.17.0.188 port 5000: Connection refused
! Check attempt 1/6 failed.
-----> Attempt 2/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/ => ""
!
curl: (7) Failed to connect to 172.17.0.188 port 5000: Connection refused
! Check attempt 2/6 failed.
-----> Attempt 3/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/ => ""
!
curl: (22) The requested URL returned error: 500 Internal Server Error
! Check attempt 3/6 failed.
-----> Attempt 4/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/ => ""
!
curl: (22) The requested URL returned error: 500 Internal Server Error
! Check attempt 4/6 failed.
-----> Attempt 5/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/ => ""
!
curl: (22) The requested URL returned error: 500 Internal Server Error
! Check attempt 5/6 failed.
-----> Attempt 6/6 Waiting for 10 seconds ...
CHECKS expected result:
http://localhost/ => ""
!
curl: (22) The requested URL returned error: 500 Internal Server Error
Could not start due to 1 failed checks.
! Check attempt 6/6 failed.
=====> myapp container output:
=> Booting Thin
=> Rails 4.2.0 application starting in production on http://0.0.0.0:5000
=> Run `rails server -h` for more startup options
=> Ctrl-C to shutdown server
Thin web server (v1.6.3 codename Protein Powder)
Maximum connections set to 1024
Listening on 0.0.0.0:5000, CTRL+C to stop
Started GET "/" for 172.17.42.1 at 2015-03-26 21:36:47 +0000
Is the server running on host "172.17.42.1" and accepting
TCP/IP connections on port 5431?
PG::ConnectionBad (could not connect to server: Connection refused
Is the server running on host "172.17.42.1" and accepting
TCP/IP connections on port 5431?
):
vendor/bundle/ruby/2.0.0/gems/activerecord-4.2.0/lib/active_record/connection_adapters/postgresql_adapter.rb:651:in `initialize'
vendor/bundle/ruby/2.0.0/gems/activerecord-4.2.0/lib/active_record/connection_adapters/postgresql_adapter.rb:651:in `new'
vendor/bundle/ruby/2.0.0/gems/activerecord-4.2.0/lib/active_record/connection_adapters/postgresql_adapter.rb:651:in `connect'
vendor/bundle/ruby/2.0.0/gems/activerecord-4.2.0/lib/active_record/connection_adapters/postgresql_adapter.rb:242:in `initialize'
vendor/bundle/ruby/2.0.0/gems/activerecord-4.2.0/lib/active_record/connection_adapters/postgresql_adapter.rb:44:in `new'
vendor/bundle/ruby/2.0.0/gems/activerecord-4.2.0/lib/active_record/connection_adapters/postgresql_adapter.rb:44:in `postgresql_connection
=====> end myapp container output
/usr/bin/dokku: line 49: 23409 Killed dokku deploy "$APP"
To dokku@dokku.example.com:myapp
! [remote rejected] dokku -> master (pre-receive hook declined)
error: failed to push some refs to 'dokku@dokku.example.com:myapp'
Configuring docker stop timeout
By default, docker will wait 10 seconds from the time the stop command is passed to a container before it attempts to kill said container. This timeout can be configured on a per-app basis in dokku by setting the DOKKU_DOCKER_STOP_TIMEOUT configuration variable. This timeout applies to normal zero-downtime deployments as well as the ps:stop and apps:destroy commands.