Multiplot/fetch_multi changes #460
Conversation
|
|
||
| if r.status_code == 200: | ||
| return [m['metric'] for m in r.json()] | ||
| else: |
There was a problem hiding this comment.
Is it worth adding retries over here?
There was a problem hiding this comment.
People have been using it without a problem I think, and I've had to work on so many other issues that I'm not going to do this one yet.
GeorgeJahad
force-pushed
the
george/add_fetch_multi2
branch
from
a0d8e81
to
c21dd40
Compare
|
|
||
| def complete(self, metric, query_depth): | ||
| metric_parts = metric.split('.') | ||
| if query_depth < len(metric_parts): |
There was a problem hiding this comment.
Totally up to you, but what do you think about consolidating these if statements into just one block like, if (query_depth < len(metric_parts)) or (self.enable_submetrics and (query_depth == len(metric_parts)))
|
These files have been updated per the review from chinmay-gupte.
Note that if you only want to see the diffs since the initial |
| # names, so return leaf nodes for each submetric alias that is | ||
| # enabled | ||
| if query_parts[-1] == '*': | ||
| for metric in self.find_metrics('.'.join(query_parts[0:-1])): |
There was a problem hiding this comment.
If the query ends with a * but its not for a leaf node, it will make a call to find_metrics here and because its not complete, it will make a call again to find_metrics to return a branch node. Are these actual API calls or the data returned is cached?
There was a problem hiding this comment.
If its doing the former, can we make just one API call in this function to find_metrics and cache the data in a local variable?
|
The lastest diffs: |
| yield BranchNode('.'.join(parts[:query_depth])) | ||
| # First modify the pattern to get a superset that includes already complete | ||
| # submetrics | ||
| new_pattern = complete_pattern + '*' |
There was a problem hiding this comment.
Is the variable new_pattern same as original query.pattern?
There was a problem hiding this comment.
no it isn't. It is a superset which is what I was trying to explain in the comment. For example, if query.pattern is "a.b.", it won't return "a.b", so new_pattern is "a.b", with no second ".". Does that make sense? I'm trying to eliminate the second api call by combining the two.
|
+1 after my last comment has been addressed. |
Multiplot/fetch_multi changes
Made a number of fixes/enhancements to the finder:
equidistant in time, (interleaved with Null datapoints if necessary.)
We weren't doing this so the timescale was off on graphs with sparse data.
(Note this required updating the graphite-api code from version 1.0.1 to
Head of master. Since we also had to patch that code I created a new repo
for it here: https://github.com/rackerlabs/graphite-api/tree/george/fetch_multi_with_patches
account. Multplot allows only 100 metrics per call, and has implicit limit of around
7000 json characters. So my changes batch up each of graphite-apis fetch_multi() calls
into to the minimum number of multiplot calls that still abide by those limits.
than just assuming "average", "num_points" or "latest" as we were before.
This required changes on both the search side, and the fetch_multi side.
This work was based on the changes suggested and created by Alex Weeks:
https://github.com/aweeks
https://23282e1499f9e238db58e26751fc6d96be63cc17-www.googledrive.com/host/0B28IWTWKORCkfnlLYmljbW1NQ295NWhWbmtfbDJ1UXYwQm9xdlVuMjN0NFV1QTNmcDltY2M/x/blueflood.html