Skip to content
Log in Register
Donate
GitHub

Monitoring

AI assistant

Build this with Claude Code, Cursor or Copilot

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

Health checks

Check Command Healthy
Installation vendor/bin/tessero doctor exit code 0
Machine-readable facts vendor/bin/tessero info --json JSON with libtessero, simd, blas, openmp
Laravel php artisan tessero:doctor --json backend is what you deployed
Extension php -r 'echo Tessero\Ext\Engine::version();' the version you deployed

Run the check in the same SAPI and php.ini as production. For FPM, expose a small internal endpoint rather than relying on the CLI:

Route::get('/internal/tessero', fn (Tessero\Laravel\TesseroManager $t) => response()->json($t->info()))
    ->middleware('internal-only');

Add a functional probe that computes something small and checks the answer, so a broken BLAS or a wrong binary is caught as well as a missing one:

$ok = Tessero::linprog([-1, -1], [[1, 2], [3, 1]], [4, 6])['success']
   && abs(Tessero::array([1.0, 2.0, 3.0])->sum() - 6.0) < 1e-12;

Metrics worth exporting

Metric Source Why
native bytes in use Engine::memoryInUse() / Tessero::memoryInUse() at request/job end should return to a baseline. A rising floor is a leak in application code (arrays held in statics)
native peak peakMemory() budget headroom. Alert when the peak approaches the budget
mapped bytes Engine::info()['memory_mapped'] file mappings held by the process; should return to 0 when memmaps go out of scope
MemoryError / MemoryException count exception handler requests hitting the budget: either raise the budget or tighten input validation
solver status counts status field of LP/MILP/MDP results a rise in 1 (limit) or 4 (numerical) means models have grown or become ill-conditioned
solve duration wrap calls in your timer capacity planning for queue workers
nodes / iterations solver results early warning before limits are hit

Example: a Laravel job middleware that records memory and duration.

final class TesseroMetrics
{
    public function handle(object $job, Closure $next): void
    {
        $t = hrtime(true);
        try {
            $next($job);
        } finally {
            Log::info('tessero.job', [
                'job'        => $job::class,
                'ms'         => (hrtime(true) - $t) / 1e6,
                'native_mb'  => Tessero\Ext\Engine::memoryInUse() / 1048576,
                'native_peak_mb' => Tessero\Ext\Engine::peakMemory() / 1048576,
            ]);
        }
    }
}

Logs

Tessero writes no logs of its own. Everything it has to report arrives as an exception or a result status. Log:

  • exception class and message. Messages include shapes, sizes and the offending state/action or constraint.
  • solver message and status for non-optimal results.
  • the backend and version once at worker start (TesseroManager::info()).

What not to alert on

  • The first request after an FPM restart being slower: preload and OpenMP pool start-up. It is a one-off per worker.
  • peakMemory() staying high: it is a high-water mark for the process lifetime and never decreases.

Edit this page on GitHub