Roland Johann created ZEPPELIN-3765:
---------------------------------------
Summary: Dependency download via Proxy inefficient
Key: ZEPPELIN-3765
URL: https://issues.apache.org/jira/browse/ZEPPELIN-3765
Project: Zeppelin
Issue Type: Bug
Components: Interpreters
Affects Versions: 0.8.0
Reporter: Roland Johann
Zeppelin tries to download dependencies with repositories in defined order:
* default maven central
* local
* custom/self defined
When zeppelin must use a http proxy custom defined maven central with proxy
properties solve this particular issue and introduces another one, based on the
ordering of dependency resolution with defined repositories: First default
maven central will be used, when the connection can't be established the used
library (sonatype aether and apache wagon) wait 120 seconds until the request
fails, then tries subsequent repositories.
In the current implementation one configures aether to use shorter timeouts
with system properties `aether.connector.connectTimeout` and
`aether.connector.requestTimeout` which will be used at
`WagonRepositoryConnector.java:310` but never will be passet to the actual http
client.
As apache wagon with `LightweightHttpWagon` implementation is in use which
itself uses plain Java `sun.net.www.http.HttpClient` properties to set timeouts
will be `sun.net.client.defaultConnectTimeout` and
`sun.net.client.defaultReadTimeout`.
These Properties can be used as workaround but we shouldn't rely on timeouts
here.
We can solve the problem in several ways:
Change the kind default repositories are treated/specified:
* ordering of repos: first custom, then default
* don't specify default repositories at code but at `interpreter.json`
configuration file
* when still definition at code, default maven central repo should get
optional proxy/auth at `org.apache.zeppelin.dep.Booter#newCentralRepository`
I prefer that there are no repo definitions at code and the user can specify
everything at `interpreter.json`.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
