Machine learning request

Error message

Machine learning request '${JSON.stringify(config)}' failed for all URLs

What it means

The machine learning repository attempts the prediction request against each configured ML URL in order, logging a warning per failed URL. If every URL fails (connection refused, timeout, HTTP errors), it throws with the serialized request config, indicating the ML service is unreachable or unhealthy rather than a problem with the request itself.

Solutions

Verify the ML service is up and reachable: curl http://machine-learning:3003/ping from the server container
Check/fix MACHINE_LEARNING_URL(S) env — it must match the ML container's host:port and Docker network
Inspect ML container logs for startup/model-load crashes and fix (e.g. increase memory, fix model name)
Ensure request payload matches the running ML server's supported schema/version (align Immich server and ML image versions)

Example fix

// before (docker-compose.yml, server env)
MACHINE_LEARNING_URL=http://127.0.0.1:3003
// after — use the compose service name
MACHINE_LEARNING_URL=http://immich-machine-learning:3003

Defensive patterns

Strategy: retry

Validate before calling

for (const url of mlUrls) {
  try { const r = await fetch(`${url}/ping`); if (!r.ok) throw new Error(String(r.status)); }
  catch { throw new Error(`ML service unreachable at ${url}; check container/network`); }
}

Type guard

function mlUrlLooksValid(url: string): boolean {
  try { const u = new URL(url); return u.protocol === 'http:' || u.protocol === 'https:'; } catch { return false; }
}

Try / catch

try {
  const result = await mlRepo.predict(url, config, input);
} catch (e) {
  if ((e as Error).message.includes('failed for all URLs')) {
    logger.warn('ML unreachable; queueing for retry', { config });
    await jobRepo.add({ name: JobName.X, data: input }); // retry via queue with backoff
  } else throw e;
}

Prevention

Health-check the ML /ping endpoint in readiness probes before serving traffic
Use compose service DNS names, not localhost, for cross-container ML URLs
Keep server and ML image versions aligned and monitor ML container logs/memory

When it happens

Trigger: All entries in MACHINE_LEARNING_URLS are unreachable at request time: ML container down, wrong host/port, network policy blocking the call, or the ML service crashing on the request — after which the loop falls through to the final throw.

Common situations: ML container not started or crashed (OOM during model load); misconfigured MACHINE_LEARNING_URL (e.g. 127.0.0.1 when ML runs in another container); Docker network misconfiguration; ML model failing on a particular image so every attempt errors.

Understand the failure class

Background: 'Something went wrong' / 'Request failed (500)' / 'HTTP error! status: 404' — what failed HTTP requests actually mean and how to find the real cause — this error's family across 28 libraries.

Related errors

AI-assisted analysis of immich-app/immich@e55ac299a4 (2026-09-15). Data as JSON: /api/errors/d6a876ffe6612af9. Report an issue: GitHub.

Appendix: source

Thrown at server/src/repositories/machine-learning.repository.ts:187

    ]) {
      try {
        const response = await fetch(new URL('predict', url), { method: 'POST', body: formData });
        if (response.ok) {
          this.setHealthy(url, true);
          return response.json();
        }

        this.logger.warn(
          `Machine learning request to "${url}" failed with status ${response.status}: ${response.statusText}`,
        );
      } catch (error: Error | unknown) {
        this.logger.warn(`Machine learning request to "${url}" failed`, error);
      }

      this.setHealthy(url, false);
    }

    throw new Error(`Machine learning request '${JSON.stringify(config)}' failed for all URLs`);
  }

  async detectFaces(imagePath: string, { modelName, minScore }: FaceDetectionOptions) {
    const request = {
      [ModelTask.FACIAL_RECOGNITION]: {
        [ModelType.DETECTION]: { modelName, options: { minScore } },
        [ModelType.RECOGNITION]: { modelName },
      },
    };
    const response = await this.predict<FacialRecognitionResponse>({ imagePath }, request);
    return {
      imageHeight: response.imageHeight,
      imageWidth: response.imageWidth,
      faces: response[ModelTask.FACIAL_RECOGNITION],
    };
  }

  async encodeImage(imagePath: string, { modelName }: MachineLearningConfig['clip']) {

View on GitHub (pinned to e55ac299a4)