We had a similar issue a while back.  I don't think we ever tracked down what 
the original cause of it was.  The code that adds in the entries is supposed to 
clean it up too (limit it to the last 10 entries).  One of the problems we ran 
into after a while was that ZK was letting the add succeed, but the listing was 
failing because there were do many entries the buffer was filling up, which was 
just making it worse.
We might want to switch the order of operations.  Only add in an entry if you 
can make room for it before.


- Bobby


On Wednesday, June 28, 2017, 2:40:29 PM CDT, Kristopher Kane 
<[email protected]> wrote:

Thanks. That is what I was looking for.  
We had a case where a fast failing storm-kafka spout (bad topic name) caused 
the creation of 600k child znodes, under one topology name, in /storm/errors in 
four days. 
On Wed, Jun 28, 2017 at 2:48 PM, Bobby Evans <[email protected]> wrote:

That depends on the version of storm you have.  In the newer versions of storm 
everything in ZK is thrift encoded objects, that are probably gziped.
https://github.com/apache/ storm/blob/master/storm- client/src/jvm/org/apache/ 
storm/cluster/ StormClusterStateImpl.java
Is how storm interacts with ZK for most things.  Please don't use it directly 
but it can give you an idea of what is stored in ZK and where.
For errors I think https://github.com/apache/ storm/blob/master/storm- 
client/src/jvm/org/apache/ storm/cluster/ StormClusterStateImpl.java# L659-L691 
is what you want to look at.

- Bobby


On Tuesday, June 27, 2017, 2:15:09 PM CDT, Kristopher Kane 
<[email protected]> wrote:

I should qualify the path there is the ZooKeeper znode for /storm/errors


On Tue, Jun 27, 2017 at 1:30 PM, Kristopher Kane <[email protected]> wrote:

Is there a doc or explanation for the contents populated under 
/storm/errors/<topology_name>
Thanks, 
Kris



Reply via email to