Github user dbrinegar commented on a diff in the pull request:
https://github.com/apache/flink/pull/4188#discussion_r124366697
--- Diff:
flink-metrics/flink-metrics-statsd/src/main/java/org/apache/flink/metrics/statsd/StatsDReporter.java
---
@@ -179,41 +254,130 @@ private String prefix(String ... names) {
}
}
- private void send(final String name, final String value) {
+ private String buildStatsdLine(final String name, final String value,
final String tags) {
+ Double number;
try {
- String formatted = String.format("%s:%s|g", name,
value);
- byte[] data =
formatted.getBytes(StandardCharsets.UTF_8);
- socket.send(new DatagramPacket(data, data.length,
this.address));
+ number = Double.parseDouble(value);
+ }
+ catch (NumberFormatException e) {
+ // quietly skip values like "n/a"
+ return "";
}
- catch (IOException e) {
- LOG.error("unable to send packet to statsd at '{}:{}'",
address.getHostName(), address.getPort());
+ if (number >= 0.) {
+ return String.format("%s:%s|g%s", name, value, tags !=
null ? tags : "");
+ } else {
+ // quietly skip "unknowns" like
lowWaterMark:-9223372036854775808, or JVM.Memory.NonHeap.Max:-1, or NaN
+ return "";
}
}
- @Override
- public String filterCharacters(String input) {
+ private void send(final String name, final String value, final String
tags) {
+ String formatted = buildStatsdLine(name, value, tags);
+ if (formatted.length() > 0) {
+ try {
+ byte[] data =
formatted.getBytes(StandardCharsets.UTF_8);
+ socket.send(new DatagramPacket(data,
data.length, this.address));
+ }
+ catch (IOException e) {
+ LOG.error("unable to send packet to statsd at
'{}:{}'", address.getHostName(), address.getPort());
+ }
+ }
+ }
+
+ /**
+ * dogstatsd names should: start with letter, uses ascii alphanumerics
and underscore, separated by periods.
+ * Collapse runs of invalid characters into an underscore. Discard
invalid prefix and suffix.
+ * Eg: ":::metric:::name:::" -> "metric_name"
+ */
+
+ private boolean isValidStatsdChar(char c) {
+ return (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || (c
>= '0' && c <= '9') || (c == '_');
--- End diff --
Yeah that's a fair point. The thought here is to take an extremely limited
set to maximize compatibility with collectors and timeseries databases, to
avoid translation. I think if one were to make a compatibliity table you might
find a few extra punctuation type characters, but I couldn't see how they
changed the significance or meaning of the metrics, so landed on picking one
commonly accepted non-period delimiter. The underbar is also guidance from
datadog, for best practice with their systems. The metrics from Flink are
super clean this way. Flink metric names seem to always be alpha and dots, the
tag names are all alpha and underscore (eg `task_id`), so looks quite natural
for the tag values to be alphanumeric + underscore.
But yeah, is an arbitrary choice.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
