Method parseRunnerLog

src/cml.js:370–425

(opts = {})

  }

  async parseRunnerLog(opts = {}) {
    let { data, name, cloudSpot } = opts;
    if (!data) return [];

    data = data.toString('utf8');

    const parseId = (key) => {
      if (patterns[key]) {
        const regex = patterns[key];
        const matches = regex.exec(data) || [];
        return matches[1];
      }
    };

    const driver = await this.getDriver();
    const logs = [];
    const patterns = driver.runnerLogPatterns();
    for (const status of ['ready', 'job_started', 'job_ended']) {
      const regex = patterns[status];
      if (regex.test(data)) {
        const date = new Date();
        const log = {
          status,
          date: date.toISOString(),
          repo: this.repo
        };

        if (status === 'job_started') {
          log.job = parseId('job');
          log.pipeline = parseId('pipeline');

          // GitHub API doesn’t seem to provide a straightforward way to get the
          // job identifier from the runner logs, so we need to query several API
          // endpoints to retrieve it. Due to several broken parts of our logic and
          // some unexpected API responses, performing these queries may trigger API
          // rate limits, causing the whole `cml` process to crash. Given that
          // retrieving the job identifier is only useful for spot instance recovery
          // (i.e. automated retryWorkflow), we can save ourselves all the hassle if
          // —-cloud-spot is not set or —-driver is not GitHub.
          if (cloudSpot && this.driver === GITHUB) {
            const { id: runnerId } = await this.runnerByName({ name });
            const { id } = await driver.runnerJob({ runnerId });
            log.job = id;
          }
        }

        if (status === 'job_ended')
          log.success = patterns.job_ended_succeded.test(data);

        log.level = log.success ? 'info' : 'error';
        logs.push(log);
      }
    }

    return logs;
  }

  async startRunner(opts = {}) {