Skip to content

Installation Intelligence (Telemetry) ​

Horilla can optionally send anonymous technical metadata so the Horilla project can see which versions and environments are in use. This helps decide which Python, Django and database versions to keep supporting.

It is opt-in and off by default. A fresh install sends nothing until a superuser explicitly enables it, and no HR data is ever transmitted.

Looking for the privacy notice?

This page is the operator guide — what is sent, and how to control or disable it. For the legal notice (data controller, legal basis, retention and your erasure rights), see the Horilla privacy notice at horilla.com/privacy.

1. What is sent ​

Only these JSON keys, once per week, and only if a superuser turns the setting on:

FieldExamplePurpose
schema_version1Payload format version
installation_idrandom UUID4Distinguishes installs so one is not counted many times
horilla_version2.0.0Version adoption
python_version3.12.0Which runtimes to keep supporting
django_version5.2Framework support window
database_enginepostgresqlWhich databases to prioritise
deploymentdockerbare, docker, compose or k8s
osLinuxPlatform support
architecturex86_64Platform support
enabled_modules["attendance", "payroll"]Which first-party modules are actually used
editioncommunityCommunity vs enterprise mix

The payload is built from a fixed allowlist and is rejected at runtime if it ever contains a key outside that list. It is capped at 8 KB.

2. What is never sent ​

No employee, payroll, leave, attendance or document data. Specifically, none of the following ever leaves your server:

  • Any HR record of any kind
  • Company name, country, org size or timezone
  • Email addresses, usernames or IP addresses
  • Hostname, database name, host or password
  • SECRET_KEY or any credential

Module names are also filtered against a fixed list of first-party apps, so a private or custom app label cannot appear in the payload.

3. Enable or disable ​

The toggle lives at Settings → Privacy → Privacy & Product Insights. A superuser can also preview the exact payload from that page before deciding.

Three environment variables control it independently of the UI:

bash
HORILLA_TELEMETRY=0          # Hard kill switch: forbids pings even if the UI toggle is on
HORILLA_VERSION_CHECK=0      # Skips version/security notices (default: already off)
HORILLA_TELEMETRY_SCHEDULER=0  # Starts no background scheduler (see section 5)

HORILLA_TELEMETRY=0 is the setting to use for an air-gapped or policy-restricted deployment: it takes precedence over the database toggle, so no ping can be sent regardless of what an admin clicks.

Installs running with DEBUG=True never send anything unless HORILLA_TELEMETRY_FORCE=1 is also set.

4. Version and security notices ​

Separately from telemetry, Horilla can check for version and security notices. This is off by default and sends no installation ID — it is an anonymous GET, not a heartbeat.

bash
HORILLA_VERSION_CHECK=1      # Enable notices

Notice links are allowlisted to horilla.com, docs.horilla.com and github.com/horilla/, and a notice can never disable your install or change a setting.

5. Process model ​

The scheduler runs in one process per machine, chosen by a file lock in the temp directory. This matters in a few deployments:

  • gunicorn --preload is handled automatically. The master defers rather than starting a scheduler thread that would never fire, and the first worker to serve a request starts it instead. Horilla's own docker/gunicorn.conf.py sets preload_app = False, so this only affects custom invocations.
  • Containers and Kubernetes pods do not share a temp directory, so the lock cannot coordinate across them. An N-pod deployment sends N heartbeats a week. Install counts stay correct — every pod reports the same database-held installation_id, and the service upserts on it — only the ping rate is multiplied. To avoid that, set HORILLA_TELEMETRY_SCHEDULER=0 on all but one replica, or on all of them plus a single CronJob.
  • Platforms without file locking (notably Windows) start no scheduler at all, rather than one per worker. Use the cron route below.

To run it from cron instead of the in-process scheduler:

bash
HORILLA_TELEMETRY_SCHEDULER=0   # in your environment
bash
python manage.py telemetry send        # weekly heartbeat
python manage.py telemetry advisories  # version notices, if enabled

manage.py telemetry preview prints the exact payload without sending it or creating an installation ID.

6. Firewall ​

Allow outbound HTTPS to telemetry.horilla.com only if you opt in. If the host is unreachable, Horilla keeps working normally — the send fails silently and is retried on the next weekly run.

The endpoint is hardcoded in the Community client; it cannot be pointed at a different URL.

7. Installation ID, backup and restore ​

The installation ID is stored in the database, so it travels with your backups rather than being tied to a container or volume.

Restoring a production dump onto staging therefore copies the ID, which would make one install look like two. Use Regenerate ID on the staging copy.

  • Regenerate ID erases the previous record on the telemetry service before minting the new ID, so the copy does not leave an orphaned record behind. If the service is unreachable at that moment, you get a warning saying the old record survived — mail support@horilla.com to have it removed.
  • Delete history erases everything reported for this installation and then discards the local ID, so a later heartbeat cannot recreate it. Product insights are switched off at the same time. If the service cannot be reached, nothing is deleted and your ID is left intact so you can retry.

Both actions tell you which of the two outcomes happened, rather than reporting success unconditionally.