centos6/RHEL6 incompatible

[tommygod3]

Describe the bug:
After installing successfully and changing my Django settings.py, python manage.py runserver fails.
Stack trace:

 File "/home/vagrant/miniconda3/envs/download-stats/lib/python3.7/site-packages/elasticapm/contrib/django/apps.py", line 30, in ready
    self.client = get_client()
  File "/home/vagrant/miniconda3/envs/download-stats/lib/python3.7/site-packages/elasticapm/contrib/django/client.py", line 53, in get_client
    instance = client_class()
  File "/home/vagrant/miniconda3/envs/download-stats/lib/python3.7/site-packages/elasticapm/contrib/django/client.py", line 69, in __init__
    super(DjangoClient, self).__init__(config, **inline)
  File "/home/vagrant/miniconda3/envs/download-stats/lib/python3.7/site-packages/elasticapm/base.py", line 148, in __init__
    self._metrics.register(path)
  File "/home/vagrant/miniconda3/envs/download-stats/lib/python3.7/site-packages/elasticapm/metrics/base_metrics.py", line 38, in register
    self._metricsets[class_path] = class_obj()
  File "/home/vagrant/miniconda3/envs/download-stats/lib/python3.7/site-packages/elasticapm/metrics/sets/cpu_linux.py", line 30, in __init__
    self.previous.update(self.read_system_stats())
  File "/home/vagrant/miniconda3/envs/download-stats/lib/python3.7/site-packages/elasticapm/metrics/sets/cpu_linux.py", line 59, in read_system_stats

    int, whitespace_re.split(line)[1:-1]
ValueError: not enough values to unpack (expected 10, got 9)

I have looked into it and the `cpu_linux.py' expects the system stats file to have 10 fields:

if line.startswith("cpu "):
                    user, nice, system, idle, iowait, irq, softirq, steal, guest, guest_nice = map(
                        int, whitespace_re.split(line)[1:-1]
                    )

However when I look at /proc/stat which is where cpu_linux.py is looking, I see it has only 9 fields:

cpu  162679 114 68061 7260831 1061 1237 1080 0 0

This has been tested on various centos6 VMs and they all had 9 fields in the system stats file. The fields are different to what cpu_linux.py expects and can be found here: https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/deployment_guide/s2-proc-stat.

To Reproduce

1. pip install elastic-apm on a centos6/rhel6 machine.
2. Configure settings.py appropriately.
3. Run python manage.py runserver.
4. Error.

Expected behavior:

Should work on centos/rhel machines and not assume all system stats files look the same.

Environment

• OS: centos6
• Python version: 3.7.1
• Framework and version: Django 2.1.2
• APM Server version: 6.6.0
• Agent version: 4.1.0

Additional context

My current workaround: override the decision to use cpu_linux.py in cpu.py by setting the environment variable ELASTIC_APM_FORCE_PSUTIL_METRICS. This uses cpu_psutil.py instead which requires you have installed psutil.

[beniwohli]

Great catch @tommygod3! I implemented a fix for this issue in #407

[tommygod3]

Awesome, cheers @beniwohli