Getting Requests per second RPS

[bab5470]

We're working with a vendor that needs to know our PowerDNS RPS. I can see at http://127.0.0.1:8081/api/v1/servers/localhost/statistics we can get stats for requests but I don't see a cumulative number anywhere. Is there any way to easily tally up the information. Or is there a metric I'm missing?

I've tried doing jq -r '.[] | select(.name == "queries") | .value | map(.value | tonumber) | add' but that gives me parse error: Invalid numeric literal at line 1, column 10

The output from the API endpoint returns almost 2,500 lines of output. Adding it up by hand just isn't reasonable. Everything is also broken up into multiple categories like serverfail-queries and unauth-queries. I'm looking for TOTAL queries (of any type) for a given moment.

Any help would be appreciated.

[bab5470]

I see this:

{
    "name": "udp-queries",
    "type": "StatisticItem",
    "value": "xxxxxx"
},

I don't see a corresponding tcp-queries though. And what timeframe is this?

This https://docs.powerdns.com/recursor/metrics.html seems to suggest that stats are output every half hour or so based on statistics-interval which is 1800 seconds (30 minutes)

We're logging to syslog though (/var/log/messages). I'm grepping that for any stats and not seeing anything:

$cat /var/log/messages | grep queries
$cat /var/log/messages | grep stats

[Habbie]

is your question about authoritative or recursor?

[bab5470]

Sorry I should have mentioned - authoritative.

[omoerbeek]

It helps to no confuse authoritaive server and recursor.

See the general API docs at https://docs.powerdns.com/authoritative/http-api/index.html, https://docs.powerdns.com/authoritative/http-api/statistics.html can be used to query specific metrics.

Many of our users use Prometheus to get metrics and compute aggregate numbers and graphs, see https://docs.powerdns.com/authoritative/http-api/index.html#metrics-endpoint

[bab5470]

Thank you but I'm not sure that helps? I'm querying http://dns1.mydomain.internal:8081/api/v1/servers/localhost/statistics?=

This is from https://docs.powerdns.com/authoritative/http-api/statistics.html.

The trouble is that it returns 2,500 lines of output broken up into hundreds, if not thousands, of categories, and I don't see any cumulative statistics. Attempts to tally up these numbers using JQ are failing, but maybe I just don't have the right combination of JQ commands.

I do see this metric which looks like it might be a cumulative number of some sort:

{
    "name": "udp-queries",
    "type": "StatisticItem",
    "value": "3069309"
},

But:

Does this include successful and unsuccessful queries or is it just successful queries?
What about TCP queries? I don't see a metric for that.
What timeframe is this? Is this live data? Queries in the last minute, last 5 minutes? Last hour? Since the server was restarted?

I'm trying to obtain requests per second... so I need live data or I need to know the interval to calculate the live interval. Say for example this was 1 hour of data I'd take 3069309/3600 seconds/hour = 852 RPS but if its 5 minutes of data then 3069309/300 second/5 minutes = 10231 RPS which is a vastly different number.

The recusor documentation mentions statistics-interval is 1800 seconds (30 minutes). Is it the same for the authoritive server as well? I don't see any reference to the statistics interval for authoritive server in the documentation. Appologies if I've missed or overlooked that somewhere.

And sorry again for mixing up the two sets of documentation.

Thanks!

[omoerbeek]

1. The docs say you can retrieve individual stats:

$ curl --silent -H 'X-API-Key: XXXX' http://localhost:8081/api/v1/servers/localhost/statistics?statistic=udp-queries | jq
[
  {
    "name": "udp-queries",
    "type": "StatisticItem",
    "value": "59"
  }
]

2. There is also a tcp-queries metric. The two counters count incoming queries, successful or not.
3. You can learn about available metrics from https://docs.powerdns.com/authoritative/http-api/index.html#metrics-endpoint. It lists the Prometheus names, the native names follow a clear pattern: strip pdns_ and replace _ by - for the native name.
4. All metrics are from server startup. You have to compute qps yourselves. You can use the uptime metric for that, giving you a qps = (udp-queries + tcp-queries) / uptime since startup. If you want to have a different period (e.q. over the last 5min) you have to do that yourself. As said, Prometheus can be of big help.
5. The recursor's logs periodic stats, auth does not. (The recursor's metrics from the API are lifetime based, just as auth's).