Extract quota config from service config.

[chowchow316]

No description provided.

[istio-testing]

Jenkins job proxy/presubmit passed

[qiwzhang]

please add unit_test

[qiwzhang]

LogError("Metric rule with selector %s is mismatched")

[chowchow316]

env->LogError() only accepts one input variable.

[qiwzhang]

quota().grouprules()

[chowchow316]

There is no grouprules field in quota: https://github.com/cloudendpoints/service-control-client-cxx/blob/rate_limiting/proto/google/api/quota.proto

[qiwzhang]

sorry, it is "rules" then

[qiwzhang]

we can speed up this code by creating a map of group_name to group proto.

[mangchiandjjoe]

one minor thing, could consider using std::unorderded_map or std::vector<std::pair<std::string, int>>.
I recommend using vector, since api_manager would not call find, it just iterates the list.

[qiwzhang]

we can use std::vector in methodinfo.

[chowchow316]

Just wondering why should we use vector instead of map here?

[qiwzhang]

if you don't need to find by key, it is better to use vector.

[mangchiandjjoe]

std::map using balanced tree to maintain order and lookup. if we don't need an item order, unordered_map is more efficient. and if we don't lookup, vector is better for iteration.

[mangchiandjjoe]

https://cs.corp.google.com/piper///depot/google3/tech/internal/env/framework/wrappers/chemist/chemist_wrapper.cc?rcl=145718473&l=725

According to the source code

• quota_metrics conversion from rules and metric_rules are different.
• if there is rules, then metric_rules will be ignored.
• cost field can be empty, it should be 1

[mangchiandjjoe]

rules {
selector: "apiserving.servicecontrol.testing.v1.QuotaTest.User"
groups {
group: "ReadGroupUser"
cost: 3
}
}

will be converted to

quota_metrics: { # came from rules setting
name: "serviceruntime.googleapis.com/api/consumer/quota_used_count"
metric_values: {
Labels: {
key: "/quota_name"
value: "ReadGroupUser"
}
value: {
Int64_value: 3
}
}
}

metric_rules {
selector: "apiserving.servicecontrol.testing.v1.QuotaTest.User"
metric_costs {
key: "chemisttest.googleapis.com/operation/read_book"
value: 100
}
}

will be converted to

quota_metrics: { # came from metric_rules
name: "chemisttest.googleapis.com/operation/read_book"
metric_values: {
value: {
int64_value: 100
}
}

[qiwzhang]

we can add this check in a separate CL

for a selector, if there is a group_rule , don't use metric_rule

[chowchow316]

Thanks for your info, and let me make it clear: if there is rules, we do not care about anything in metric_rules, right? And we are always saving the quota info into a <metric, cost> pair, what should we use for the metric names for rules & metric_rules? It seems they have different structures.

[mangchiandjjoe]

We need to know the rule is came from rules or metric_rules. We can use the same data structure but additional information is required.

[qiwzhang]

Maybe we can do unit_test later. Let us get this clean up and check it in so Jae can start to use it.

then use separate PR to add unit_test

[qiwzhang]

call it group_map

[qiwzhang]

Just talked to Chengliang, for ESP, we don't need to worry about group rules in "rules" field. we only need to add metric_rules()

[chowchow316]

Will add unit test in another PR.

[istio-testing]

Jenkins job proxy/presubmit passed

[qiwzhang]

could you log rule.selector() too?

[qiwzhang]

in this error case, return false.

[chowchow316]

Done.

[qiwzhang]

you need to call this function in ::Create() function. it can be the last one.

[chowchow316]

Done.

[istio-testing]

Jenkins job proxy/presubmit passed