Skip to content
Log in Register
Donate
GitHub

Overview

AI assistant

Build this with Claude Code, Cursor or Copilot

Copy a Tessero-tuned prompt for Overview, grounded in 16 real API signatures , into your IDE's AI. No chatbot, just exact context.

This manual is for the people who install, run and support Tessero in production: platform engineers, SREs, and developers who own a service end to end. It assumes you know how PHP runs in your environment (FPM, Octane, queue workers, CLI).

Page Read it when
[Deployment(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/deployment.md) Putting Tessero on a server for the first time: CLI, FPM with preload, Octane, queue workers
[Native extension(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/native-extension.md) Building, packaging and rolling out ext-tessero
[Docker(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/docker.md) Building images (Debian and Alpine), multi-stage builds
[Configuration reference(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/configuration-reference.md) Every INI setting, environment variable and config key in one place
[Memory and threads(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/memory-and-threads.md) Sizing workers, setting the native memory budget, choosing thread counts
[Monitoring(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/monitoring.md) What to measure, health checks, logs
[Performance tuning(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/performance-tuning.md) A workload is slower than expected
[Troubleshooting(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/troubleshooting.md) An error message or symptom, and what to do about it
[Runbooks(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/runbooks.md) Step-by-step responses to incidents
[Security(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/security.md) Hardening, untrusted input, reporting vulnerabilities
[Upgrading(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/upgrading.md) Moving between versions
[Soak testing(https://github.com/DMJ-CV-91913/tessero/blob/main/docs/operations/soak-testing.md) Proving that long-lived workers do not accumulate memory, mappings or descriptors

The operating model in one page

  • Two memory pools. PHP's memory_limit covers PHP values. Array data lives in native memory, allocated by libtessero and not counted by memory_limit. Each process has its own counter and optional budget (memory_budget). Set the budget; it turns a runaway allocation into an exception instead of an OOM kill.
  • Memory is freed deterministically. A native block is freed when the last PHP object that references it is destroyed. There is no separate garbage collector and no finaliser queue.
  • Threads are opt-in. Kernels run on one thread unless you raise threads. Under FPM, parallelism comes from having many workers. Raise threads for CLI jobs and dedicated worker pools.
  • Settings are per process. Threads, budget and tolerance are process-wide. The Laravel provider re-applies them per request and per job. Outside Laravel, set them at bootstrap.
  • Two backends, identical numbers. The FFI package needs ffi.enable (and preload under FPM). The extension needs extension=tessero. Either can be rolled back to the other without changing results.
  • No network, no files, no background threads of its own. Tessero does not open sockets, does not read files except the ones you pass to Npy::load, and does not keep threads alive between calls, apart from the OpenMP runtime's pool when threads > 1.

Edit this page on GitHub