For the complete documentation index, see llms.txt. This page is also available as Markdown.

execserver on another linux

How to spread one fylr over two linux servers

This is a variation of the main installation method described here. So we will concentrate on the differences.

Goal

  1. One server with only the "execserver" part of fylr, which will be doing the heavy lifting of processing assets, for example generating preview thumbnails. So it should have some CPU cores. We call this one exec.example.com, here.

  2. Another server with all the remaining parts of fylr, doing Webfrontend, SQL, Indexing, etc.. We call this one main.example.com, here.

On the system with the execserver

(We call this system exec.example.com)

  • ensure that only main.example.com is allowed to reach port 8083 on exec.example.com (firewall, or private IP address, etc.)

  • use a fylr.yml like this: (which disables all parts except execserver)

fylr+:
  logger+:
    timeFormat: "2006-01-02 15:04:05Z07"

  db:

  plugin:

  elastic:

  execserver: # no need to configure the client here, as this fylr is the execserver

  services+:
    execserver+: # this is the execserver
      # all: delete failed jobs, for preserving disk space. done: keep failed jobs, for debugging
      jobRemovalPolicy: done
      addr: :8083

    api:

    backend:

    webapp:
  • use such a docker-compose.yml : (no postgreSQL, no Indexer)

  • About backups: You will have no permanent data on this system, like: no assets, no database. Just a bit of configuration.

On the system without the execserver

(We call this system main.example.com)

  • ensure that only exec.example.com is allowed to reach port 8080 and 8081 on main.example.com (firewall, or private IP address, etc.)

  • use a fylr.yml with these changes, the rest remains as in the default installation:

There is no worker count to configure on main.example.com: from 6.35 one file dispatcher sizes itself from the slots of the connected execserver, and the execserver sizes its pool to its own CPUs (slots, see performance tuning).

connectTimeoutSec is how long a job waits for a free slot, not a connection timeout. A file job that gets no slot in that time goes back into the file queue and is tried again a minute later, as often as it takes; a request whose client waits for the job, such as a plugin callback, fails with an error instead.

callbackBackendOwnURL is the backend address the execserver calls back on, callbackApiOwnURL the API address handed to plugins. Each is taken verbatim, port included, and sets its own address only. fylr 6.35.0 does not know callbackApiOwnURL yet, reports it as an unknown key and takes the API host from callbackBackendOwnURL. Before 6.35 these addresses were set with callbackBackendInternalURL and callbackApiInternalURL, which 6.35.1 ignores.

Last updated