Ovh Sdk Laravel Package

Getting Started

1. Install the package and publish its config:

   composer require ovhcloud/ovh-php-sdk   # or the Laravel wrapper, if used
   php artisan vendor:publish --tag=ovh-sdk-config
   

2. Fill in the four credentials the SDK needs — application_key, application_secret, consumer_key, and endpoint (ovh-eu, ovh-ca, ovh-us, …). Authentication is handled for you: on first call the SDK hits GET /auth/time, computes a clock delta, caches it for an hour, and signs every subsequent request.
3. Look first at the errors and DNS zones docs — they define the exception hierarchy and the task-filtering rules introduced in 0.5.0.
4. First real use case: list pending DNS-zone tasks for a domain.

   use OvhSdk\Enums\TaskStatus;
   
   $ids = $client->dns()->zone('example.com')->tasks()->ids([
       TaskStatus::Todo,
       TaskStatus::Doing,
   ]);
   

Implementation Patterns

• Resolve the client from the container and inject it. Don't new the client in controllers; the auth time-delta cache is per-instance and is worthless if you rebuild the client on every request.
• Task filtering with ids() / list() is whitelist-oriented. Pass only statuses OVHcloud will actually accept; build the array from TaskStatus cases and filter out anything you aren't sure about. These methods now validate the filter client-side before sending, so a bad filter fails fast and locally.
• Error handling — catch in the right order. The hierarchy is CouldNotDecodeResponse (subclass) → CouldNotReadResponse. Put the more specific catch first:

  try {
      $tasks = $client->dns()->zone($zone)->tasks()->list($statuses);
  } catch (CouldNotDecodeResponse $e) {
      // body wasn't JSON (empty body included) — often transient, retry
      $previous = $e->getPrevious(); // JsonException, if any
      throw $e; // or retry with backoff
  } catch (CouldNotReadResponse $e) {
      // valid JSON, wrong shape — a bug or API contract change, do not retry blindly
      report($e);
      throw $e;
  }
  

• Retry policy by exception type. CouldNotDecodeResponse is retryable (empty/HTML body from a proxy or hiccup); CouldNotReadResponse is not (scalar where a list/object was expected means the contract changed). Wire this into a queued job's retryUntil/backoff rather than a blanket catch (\Throwable).
• Handle TaskStatus::Problem on the read path. You cannot filter for it, but a task returned by OVHcloud can still carry it. Map it in your domain model and surface it in monitoring — don't assume the enum case is unreachable.
• Testing. Fake the HTTP layer (Saloon's mock / Http::fake) and assert on the SDK exceptions, not raw JsonException or TypeError. This keeps tests aligned with the new contract.

Gotchas and Tips

• Catch order is not cosmetic. CouldNotDecodeResponse extends CouldNotReadResponse; a catch (CouldNotReadResponse) placed first swallows the decode case and you lose the ability to retry only transient failures.
• The old JsonException catch is dead code. A non-JSON successful body no longer throws JsonException; it throws CouldNotDecodeResponse. Reach the underlying JsonException via getPrevious().
• Empty bodies are errors now. A list route returning an empty body used to silently become []; it now throws CouldNotDecodeResponse. Any code that treated "no content" as "no results" will start failing — decide explicitly whether to retry.
• Object route returning a JSON list now says "response is not an object" instead of naming the first missing field. Update log-parsing/greps that relied on the old wording.
• tasks()->ids() / list() reject TaskStatus::Problem with CouldNotFilterZoneTasks (an InvalidArgumentException) before the request is sent. OVHcloud's schema lists the value but answers 400 Invalid inputs — so this is a client-side guard, not a server error. Don't catch BadRequestException for it anymore.
• Auth time failures are now loud. If GET /auth/time doesn't return a JSON integer, the SDK throws instead of falling back to 0. The old fallback produced a wrong delta that was cached for an hour, so every signature failed — if you see a burst of signature errors after a deploy, check that endpoint before blaming credentials.
• Endpoint config matters. Time-delta caching is keyed per client; switching endpoint at runtime (e.g. multi-region jobs) means a fresh client and a fresh time fetch.