manage.get.gov/docs/operations

Operations

Some basic information and setup steps are included in this README.

Instructions for specific actions can be found in our runbooks.

Continuous Delivery

We use a cloud.gov service account to deploy from this repository to cloud.gov with a SpaceDeveloper user.

Authenticating to Cloud.gov via the command line

You'll need the Cloud Foundry CLI.

We use the V7 Cloud Foundry CLI.

cf login -a api.fr.cloud.gov --sso

After authenticating, make sure you are targeting the correct org and space!

cf spaces
cf target -o <ORG> -s <SPACE>

Database

In sandbox, created with cf create-service aws-rds micro-psql getgov-ENV-database.

Binding the database in manifest-<ENVIRONMENT>.json automatically inserts the connection string into the environment as DATABASE_URL.

Cloud.gov RDS documentation.

Deploy

We have three types of environments: developer "sandboxes", staging and stable. Developers can deploy locally to their sandbox whenever they want. However, only our CD service can deploy to staging and stable, and it does so when we make tagged releases of main. For staging, this is done to ensure there is a non-production level test envirornment that can be used for user testing or for testing code before it is pushed to stable. Staging can be especially helpful when testing database changes or migrations that could have adververse affects in stable. On the other hand, stable is used to ensure that we have a "golden" environment to point to. We can refer to stable as our production environment and staging as our pre-production (pre-prod) environment. As such, code on main should always be tagged for staging before it is tagged for stable.

You should make sure all of the USWDS assets are compiled and collected before deploying to your sandbox. To deploy locally to sandbox:

For ease of use, you can run the deploy.sh <sandbox name> script in the /src directory to build the assets and deploy to your sandbox. Similarly, you could run build.sh <sandbox name> script to just compile and collect the assets without deploying.

Your sandbox space should've been setup as part of the onboarding process. If this was not the case, please have an admin follow the instructions here.

Stable and Staging Release Rules

Releases will be made for staging and stable every week starting on the first day of the sprint (Wednesday), with the second release of the sprint occuring halfway through the sprint. With the exception of first time going into production, these releases will NOT have the same code. The release to stable will be the same commit that was tagged for staging one week prior, making stable one week behind staging. Further, this means staging can be up to a week behind the main branch of code.

If a bug fix or feature needs to be made to stable out of the normal cycle, this can only be done at the product owner's request.

Making bug fixes on stable during production

In the case where a bug fix or feature needs to be added outside of the normal cycle, the code-fix branch and release will be handled differently than normal:

1. Code will need to be branched NOT off of main, but off of the same commit as the most recent stable commit. This should be the one tagged with the most recent vX.XX.XX value.
2. After making the bug fix, the approved PR branch will not be merged yet, instead it will be tagged with a new release tag, incrementing the patch value from the last commit number.
3. If main and stable are on the the same commit then merge this branch into staging using the staging release tag (staging-).
4. If staging is already ahead stable, you may need to create another branch that is based off of the current staging commit, merge in your code change and then tag that branch with the staging release.
5. Wait to merge your original branch until both deploys finish. Once they succeed then merge to main per the usual process.

Serving static assets

We are using WhiteNoise plugin to serve our static assets on cloud.gov. This plugin is added to the MIDDLEWARE list in our apps settings.py.

Note that it’s a good idea to run collectstatic locally or in the docker container before pushing files up to your sandbox. This is because collectstatic relies on timestamps when deciding to whether to overwrite the existing assets in /public. Due the way files are uploaded, the compiled css in the /assets/css folder on your sandbox will have a slightly earlier timestamp than the files in /public/css, and consequently running collectstatic on your sandbox will not update public/css as you may expect. For convenience, both the deploy.sh and build.sh scripts will take care of that.

Debugging

Debugging errors observed in applications running on Cloud.gov requires being able to see the log information from the environment that the application is running in. There are (at least) three different ways to see that information: Cloud.gov dashboard, CloudFoundry CLI application, and Cloud.gov Kibana logging queries. There is also SSH access into Cloud.gov containers and Github Actions that can be used for specific tasks.

Cloud.gov dashboard

At https://dashboard.fr.cloud.gov/applications there is a list for all of the applications that a Cloud.gov user has access to. Clicking on an application goes to a screen for that individual application, e.g. https://dashboard.fr.cloud.gov/applications/2oBn9LBurIXUNpfmtZCQTCHnxUM/53b88024-1492-46aa-8fb6-1429bdb35f95/summary. On that page is a left-hand link for "Log Stream" e.g. https://dashboard.fr.cloud.gov/applications/2oBn9LBurIXUNpfmtZCQTCHnxUM/53b88024-1492-46aa-8fb6-1429bdb35f95/log-stream. That log stream shows a stream of Cloud.gov log messages. Cloud.gov has different layers that log requests. One is RTR which is the router within Cloud.gov. Messages from our Django app are prefixed with APP/PROC/WEB. While it is possible to search inside the browser for particular log messages, this is not a sophisticated interface for querying logs.

CloudFoundry CLI

When logged in with the CloudFoundry CLI (see above) Cloudfoundry application logs can be viewed with the cf logs <application> where <application> is the name of the application in the currently targeted space. By default cf logs starts a streaming view of log messages from the application. It appears to show the same information as the dashboard web application, but in the terminal. There is a --recent option that will dump things that happened prior to the current time rather than starting a stream of the present log messages, but that is also not a full log archive and search system.

CloudFoundry also offers a run-task command that can be used to run a single command in the running Cloud.gov container. For example, to run our Django admin command that loads test fixture data:

cf run-task getgov-{environment} --command "./manage.py load" --name fixtures--users
cf run-task getgov-{environment} --command "./manage.py load" --name fixtures--applications

However, this task runs asynchronously in the background without any command output, so it can sometimes be hard to know if the command has completed and if so, if it was successful.

Cloud.gov Kibana

Cloud.gov provides an instance of the log query program Kibana at https://logs.fr.cloud.gov. Kibana is powerful, but also complicated software that can take time to learn how to use most effectively. A few hints:

• Set the timeframe of the display appropriately, the default is the last 15 minutes which may not show any results in some environments.

• Kibana queries and filters can be used to narrow in on particular environments. Try the query @source.type:APP to focus on messages from the Django application or @cf.app:"getgov-{environment}" to see results from a single environment.

Currently, our application emits Python's default log format which is textual and not record-based. In particular, tracebacks are on multiple lines and show up in Kibana as multiple records that are not necessarily connected. As the application gets closer to production, we may want to switch to a JSON log format where errors will be captured by Kibana as a single message, however with a slightly more difficult developer experience when reading logs by eyeball.

SSH access

The CloudFoundry CLI provides SSH access to the running container of an application. Use cf ssh <application> to SSH into the container. To make sure that your shell is seeing the same configuration as the running application, be sure to run /tmp/lifecycle/shell very first.

Inside the container, the python code should be in /app and you can check there to see if the expected version of code is deployed in a particular file. There is no hot-reloading inside the container, so it isn't possible to make code changes there and see the results reflected in the running application. (Templates may be read directly from disk every page load so it is possible that you could change a page template and see the result in the application.)

Inside the container, it can be useful to run various Django admin commands using ./manage.py. For example, ./manage.py shell can be used to give a python interpreter where code can be run to modify objects in the database, say to make a user an administrator.

Github Actions

In order to allow some ops activities by people without CloudFoundry on a laptop, we have some ops-related actions under https://github.com/cisagov/getgov/actions.

Migrate data

This Github action runs Django's manage.py migrate command on the specified environment. This is the first thing to try when fixing 500 errors from an application environment. The migrations should be idempotent, so running the same migrations more than once should never cause an additional problem.

Reset database

Very occasionally, there are migrations that don't succeed when run against a database with data already in it. This action drops the database and re-creates it with the latest model schema. Once launched, this should never be used on the stable environment, but during development, it may be useful on the various sandbox environments. After launch, some schema changes may take the involvement of a skilled DBA to fix problems like this.

Bug triage

Bugs on production software need to be documented quickly and triaged to determine if fixes need to be made outside of the normal release cadence. Triage levels will be Critical, High, Medium, and Low to indicate the level of priority for fix, not neccessarily the level of severity. See below for more details

Critical- should only be determined by the product owner and means the fix for this critical bug needs to have a quick fix for it created ASAP. This is the only case where a bug fix can be added outside of the normal release cycle and directly onto the stable release. High- Can be determined by product owner or other team member, and indicates this bug is critical enough to warrant being added into the current sprint. Medium- Should be added to a sprint coming up but is not blocking users, or enough users to warrant rushing it into a sprint Low- A minor bug, that could even wait until after the next big launch date to be implemented.

Steps for Triaging

1. When a bug is found, whether by a developer/designer or from feedback from an end user, a ticket should be made immediately. The actual maker of the ticket can be a member of the product team as needed.
2. This bug ticket immediately gets a priority added Critical/High/Medium/Low, with Critical requiring the product owner's consent.
3. Anything marked as critical should be refined immediately and engineering should be notified in our Slack dev channel that a Critical ticket has been created (if not already notified)
4. All items not marked as critical by the product owner can wait until refinement to be refined and may have their prioirty level changed during that meeting.

Steps for dealing with Critical Bugs

1. Once the critical bug ticket is refined and the bug is clear, an engineer should be assigned to work on it. (No ticket, no work)
2. At the same point, two other engineers should be assigned to review the PR once it's made. One of the reviewing engineers can be subsititued for a designer if this is a design/content/other user facing bug fix.
3. In the case where the engineering lead is is unresponsive or unavailable to assign the ticket immediately, the product team will make sure an engineer volunteers or is assigned to the ticket/PR review ASAP.
4. Once done, the developer must make a PR and should tag the assigned PR reviewers in our Slack dev channel stating that the PR is now waiting on their review. These reviewers should drop other tasks in order to review this promptly.
5. See the the section above on Making bug fixes on stable for how to push changes to stable once the PR is approved