Custom Collector with timestamp producing exposition prometheus cannot parse

[sedan07]

I have an external system which doesn't directly expose metrics in prometheus format. To get these metrics into prometheus I'm attempting to write a custom collector in Python which scrapes the external system and publishes them in prometheus format. However the metrics from the external system are delayed by 30 minutes. Looking at the source code for the GaugeMetricFamily class it's possible to pass a timestamp. This also matches the docs on the exposition format.

However passing a timestamp causes the Prometheus server (v2.5.0) not to be able to read the metrics and gives the error: 'expected next entry after timestamp, got "MNAME"'. The following code shows the issue:

Using v0.4.2 of the python client and Python 3.7

from datetime import datetime, timedelta
import time
from prometheus_client import start_http_server
from prometheus_client.core import GaugeMetricFamily, REGISTRY

class CustomCollector(object):
    def collect(self):
        now = datetime.now()
        metric_time = now - timedelta(minutes=30)

        g = GaugeMetricFamily('my_gauge', 'Help text', labels=['foo'])
        g.add_metric(['bar'], 1.7, metric_time.timestamp())
        g.add_metric(['baz'], 3.8, metric_time.timestamp())
        yield g

def main():
    start_http_server(8000)
    REGISTRY.register(CustomCollector())

    while True: time.sleep(1)

main()

Viewing the output of localhost:8000 appears to show the metrics in the correct format matching the exposition format and doesn't show any extra values which might cause it. Am I simply using this wrong? or is this a bug somehow?

[brian-brazil]

This is a bug on the Prometheus side, it's not negotiating openmetrics correctly if gzip is not in use.