Architecture overview¶
QFieldCloud has a containerized architecture with multiple containers and volumes.
See an interactive version of the drawing above.
Note
For simplicity and clarity all graphs show the so called "happy path" without detailing the error handling through the process.
Docker objects¶
Containers¶
[nginx]
Reverse proxy¶
The reverse proxy that sits in front of QFieldCloud.
Hard requirement on nginx
for this purpose as the X-Accel-Redirect
HTTP header is heavily used in production to serve files directly from the [minio] File Storage.
Requires a SSL certificate to be present - self-signed, Let's Encrypt or other.
[app]
QFieldCloud App¶
The main software that runs QFieldCloud, including authentication, permissions management, data models, static files, administrative interface and more.
It consists of multiple Django apps, namely:
authentication
- responsible for API authentication.core
- data models, permissions, job queue management, REST API, administrative interface, and everything else in QFieldCloud that is not managed by the other Django Apps.filestorage
- responsible for project file management and respective REST API.notifs
- notifications module.subscription
- management of user's plans, subscription and storage quota.
[memcached]
In-memory cache¶
Uses memcached
for in-memory cache of Django settings and other computed values.
[ofelia]
CRON runner¶
Uses ofelia
for CRON runner.
[mkcert]
Self-signed certificate creator¶
Automatically create self-signed SSL certificate for local development and test deployments.
[certbot]
Letsencrypt manager¶
Uses certbot
for managing Let’s Encrypt certificates.
It automatically recreate expiring SSL certificate and they are automatically reloaded by [nginx
] Reverse Proxy.
[worker_wrapper]
Queue Consumer¶
One or more containers to consume and manage Jobs from the queue.
The [worker_wrapper] Queue Consumer regularly polls the [db] App Database for new pending Jobs.
Once a Job in PENDING
status is encountered, the [worker_wrapper] Queue Consumer sets it to QUEUED
status and starts processing it.
Then the container will set a bunch of metadata on the Job object in the [db] App Database and start a completely new temporary [qgis] Worker container.
It waits for the [qgis] Worker container to finish, gets the logs and stores them in the [db] App Database.
Finally, it updates a bunch of other Job's metadata in the [db] App Database based on the exit code and the logs from the [qgis] Worker, and sets the Job's status to FINISHED
or FAILED
.
The container runs the very same Django code used in [app
] QFieldCloud App.
Read how the [worker_wrapper]
Queue Consumer works in the Job Queue section.
[qgis]
Worker¶
The Worker container is the actual place where the Jobs created by the [app
] QFieldCloud App are executed.
Each Worker container is created dynamically by the [worker_wrapper
] Queue Consumer to execute a single job for a single project and deleted right after.
The typical execution of a Worker consists of starting a headless QGIS application, downloading project files, process the project files and reuploading the resulting processed files.
The [qgis]
Worker container gets environment variables, including a temporary QFieldCloud authentication token, from the [worker_wrapper]
Queue Consumer, and send back data to the it by writing logs to the STDERR
and writing files in the temporary shared volume, primarily feedback.json
.
Volumes¶
[cerbot_www]
¶
Stores the SSL certificates created by [mkcert]
Self-signed certificate creator or [certbot]
Letsencrypt manager.
[transformation_grids]
¶
Contains all transformation grids available, downloaded from https://cdn.proj.org/ and made available for all Jobs.
[static_volume]
¶
Contains all static assets such as fonts, CSS, JS or image files.
[media_volume]
¶
Warning
Deprecated, might be removed at any time.
The default location for User uploaded files. It should never be used.
Development services¶
The following containers are available only for local development purposes and should not be used in production, as these services are not monitored, backed-up and generally designed for critical usage within the stack.
[minio
] File Storage¶
Local Object Storage (S3-like) used for development.
The data is stored on 4 [minio_data
] volumes, as replication is enforced by minio
.
Should be replaced by a proper S3-like Object Storage SaaS provider.
[createbuckets
] Create Minio Buckets¶
Single shot container to create the required buckets on the Object Storage under [minio
] File Storage.
[db
] App PostgreSQL¶
Local PostgreSQL database server to host the data for the [app
] QFieldCloud App.
The database schema is entirely managed by Django.
The schema should not be modified by external services.
Should be replaced by a proper PostgreSQL database with PostGIS installed SaaS provider.
[smtpdev
] Mailing Server¶
Local mailing server to handle emails being sent, such as registration activation, reset password or notifications.
Should be replaced by a proper email SaaS provider that supports SMTP protocol.
[geodb
] GeoDB PostgreSQL¶
Warning
Deprecated, might be removed at any time.
Stores dynamically created user PostGIS databases.
[minio_data
]¶
Stores data for [geodb
] GeoDB PostgreSQL.
[postgres_data
]¶
Stores data for [db
] App PostgreSQL.
[smtp4dev_data
]¶
Stores data for [smtpdev
] Mailing Server.
[geodb_data
]¶
Warning
Deprecated, might be removed at any time.
Stores data for [geodb
] GeoDB PostgreSQL.
Job Queue¶
QFieldCloud runs very heavy operations to make Project synchronization work smoothly for users. Therefore such operations are executed using the Jobs that run in the background.
Here we have a sequence diagram of the stages each Job run follows.
Some of the steps are intentionally shortened for clarity.
sequenceDiagram
autonumber
participant DB as QFieldCloud DB
participant Queue as QFieldCloud Queue
loop poll database
Queue->>DB: select the oldest Job with `PENDING` status
Queue->>DB: set the status of Job to `STARTED` status
Queue->>DB: create a new temporary authentication token
create participant Worker
Queue->>Worker: create the worker and pass the authentication token
note over Worker: job specific processing...
destroy Worker
Queue-xWorker: get worker's structured and free text logs and exit code
Queue->>DB: set job's structured logs, free text logs and exit code
alt Worker has zero exit status
Queue->>DB: set the Job status to `FINISHED`
else Worker has non-zero exit status
Queue->>DB: set the Job status to `FAILED`
end
break when any of the steps above fails
Queue-->DB: set the status Job to `FAILED`
end
end
While some of the steps are still simplified, this diagram contains much more information about the Job's attribute lifecycle.
sequenceDiagram
autonumber
participant DB as QFieldCloud DB
participant Queue as QFieldCloud Queue
loop [every few seconds]
Queue->>DB: select the oldest job with `PENDING` status
Queue->>DB: set the status of current job to `QUEUED`
Queue->>DB: set the status of current job to `STARTED` and `started_at` to `now()`
Queue->>DB: create a new temporary authentication token
Queue->>DB: set `docker_started_at` to `now()`
create participant Worker
Queue->>Worker: - mount `transformation_grids` volume <br> - set shared data volume <br> - set the environment variable with the authentication token <br> - set the environment variables from secrets <br> - set the `pgservice` file from secrets <br><br> Create the worker
activate Worker
par run worker
Worker->>API: Download files
API->>Worker: Files downloaded
note over Worker: job specific processing...
Worker->>API: Upload files
API->>Worker: Files uploaded
deactivate Worker
and wait for Worker to finish on Queue
Queue->>DB: set `container_id` to the newly created Docker container id
Queue->>Queue: wait until the Worker is finished or timeout reached
end
Queue->>DB: set `docker_finished_at` to `now()`
destroy Worker
Queue-xWorker: - get Worker's `stdout` logs <br> - get Worker's structured logs from the `feedback.json` in the shared volume <br> - get Worker's exit status code <br><br> Remove the container and cleanup all Job data
Queue->>DB: set job's `output` and `feedback`
alt [Worker has zero exit status]
Queue->>DB: set the status of current job to `FINISHED` and `finished_at` to `now()`
else [Worker has non-zero exit status]
Queue->>DB: set the status of current job to `FAILED`
end
break when any of the steps above fails
Queue-->DB: set the status of the current job to `FAILED`
end
end
participant API as QFieldCloud App