Compatibility with graphite-web 1.1.x?

[reyjrar]

I know Booking was stuck on graphite-web 0.9.x for a long time. Have you upgraded to 1.1.x?

I ask, because while using this fork's carbonapi/carbonzipper as the CLUSTER_SERVERS for graphite-web-1.1.8, I'm not able to make any queries with globs work.

Setup:

• carbonapi listening on localhost:8082
• carbonzipper listening on localhost:8081

In graphite-web's local_settings.py

# ------------------------------------------------------------------------
# Connect to the local carbonapi endpoint
# ------------------------------------------------------------------------
CLUSTER_SERVERS = ["127.0.0.1:8081"]
INTRACLUSTER_HTTPS         = False
REMOTE_FETCH_TIMEOUT       = 90.0
REMOTE_FIND_TIMEOUT        = 30.0
REMOTE_RETRY_DELAY         = 30
REMOTE_EXCLUDE_LOCAL       = False
REMOTE_STORE_MERGE_RESULTS = False

Regardless whether this is set to 8081 (zipper) or 8082 (carbonapi), I'm able to resolve find calls, and fully qualified render targets (i.e., foo.bar.baz), but any target with a glob in it is ignored.

Like if I call /render/?format=json&target=foo.bar.baz&target=foo.*.* I get back

[ 
  { 
    "target": "foo.bar.baz",
    "tags": { ... },
    "datapoints": [ .. ]
  }
]

But if I query find for foo.*.* I get all the results I expect.

I've tried this with graphite09compat: true|false in zipper and sendGlobsAsIs: true|false in api, and every permutation thereof.

[reyjrar]

As a work-around, I spun up the python graphite-web on the data nodes and pointed CLUSTER_SERVERS to that.. it's significantly slower than using carbonapi/carbonzipper but it works and only humans are using the graphite-web end points.. I cut-over all our automation to using carbonapi.

I can also confirm that this fork can handle 10-20 times the concurrent query load as the mainline carbonapi 0.15.4.