metrics labels can Support Chinese?

[najun79]

takeout_requests_total{instance="",job="takeout_requests_total",mealscount="0",mealsname="��������",userid="user",username="������"} 3
takeout_requests_total{instance="",job="takeout_requests_total",mealscount="0",mealsname="�ϱ���ը����",userid="user",username="������"} 3
takeout_requests_total{instance="",job="takeout_requests_total",mealscount="1",mealsname="��������",userid="user",username="������"} 3
takeout_requests_total{instance="",job="takeout_requests_total",mealscount="1",mealsname="�ϱ���ը����",userid="user",username="������"} 6

labels mealsname is Chinese（utf8），takeout_requests_total can be send to pushgateway and Correctly received，but prom/Prometheus can not pull it.metrics labels can Support Chinese?

[songjiayang]

check your db charset ?

[najun79]

utf-8

[beorn7]

Label values fully support UTF-8. If you run into any issues, a detailed error report would be very helpful.

[aecolley]

Github comments seem to support 正體字 / 简化字, but the text in the mealsname of the original post appears to be gibberish (mostly Unicode "replacement characters", with the occasional Greek or Armenian letter). This makes me doubt that the text is really UTF-8. It's quite possible that the data are being passed over the net in the wrong charset, of course. Can you get intelligible responses from the pushgateway's own /metrics?

There's always the possibility that the characters being replaced by � are outside the Unicode BMP, requiring 4-byte UTF-8 encodings. That's probably not a well-tested case. But, of course, it should be.

I notice that a bare request to Prometheus's own /metrics produces a response with the header Content-Type: text/plain;version=0.0.4 without a charset parameter; a correct HTTP/1.1 client will default to ISO-8859-1. I can't immediately tell whether the bytes are ISO-8859-1 or UTF-8, because apparently "prométheus" is not a valid job name. So, this is worth further investigation IMHO.

[beorn7]

I notice that a bare request to Prometheus's own /metrics produces a response with the header Content-Type: text/plain;version=0.0.4 without a charset parameter; a correct HTTP/1.1 client will default to ISO-8859-1. I can't immediately tell whether the bytes are ISO-8859-1 or UTF-8, because apparently "prométheus" is not a valid job name. So, this is worth further investigation IMHO.

That's a good point. If the text format is used, we should definitely specify the charset in the content type. I filed prometheus/docs#557 about it.

However, in this particular case, I don't think it explains the issue. First, a charset header is not required by the protobuf format. If the UTF-8 string makes it onto the Pushgateway, it will be scraped correctly by the Prometheus server (as it uses the protobuf format). Second, if I'm not missing anything, the Prometheus server will always interpret the payload of an HTTP request as UTF-8, no matter what might or might not be in the header.

Perhaps the problem happens in the last step, namely delivering the query result via the HTTP API?

[brian-brazil]

@najun79 can you confirm if the issue is only in the HTTP API?

[brian-brazil]

No response, presuming this issue isn't with Prometheus.

[codwu]

if label value is chinese, prometheus give an error message: invalid label value "\xd5\xd7\xcd\xf8\xc2"

[brian-brazil]

This is not valid utf-8: https://play.golang.org/p/NAcJaCo7Pu

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.