[
https://issues.apache.org/jira/browse/SENTRY-296?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16022138#comment-16022138
]
Alexander Kolbasov commented on SENTRY-296:
-------------------------------------------
I looked at this implementation and I don't understand how it can work.
The usual pattern for clients is:
1) Create a new connection using SentryPolicyServiceClient.create()
2) Issue a single call (or may be a few calls)
3) Close the client.
The create() method above creates a new instance of the
PoolClientInvocationHandler:
{code}
public static SentryPolicyServiceClient create(Configuration conf) throws
Exception {
boolean haEnabled = conf.getBoolean(ClientConfig.SERVER_HA_ENABLED, false);
boolean pooled = conf.getBoolean(ClientConfig.SENTRY_POOL_ENABLED, false);
if (pooled) {
return (SentryPolicyServiceClient) Proxy
.newProxyInstance(SentryPolicyServiceClientDefaultImpl.class.getClassLoader(),
SentryPolicyServiceClientDefaultImpl.class.getInterfaces(),
new PoolClientInvocationHandler(conf)); // new instance
{code}
This creates a new pool for each connection which doesn't make any sense - the
whole purpose is to have a pool of connections shared *between* calls.
Can someone clarify how this was supposed to work?
> Sentry Service Client does not allow for connection pooling
> -----------------------------------------------------------
>
> Key: SENTRY-296
> URL: https://issues.apache.org/jira/browse/SENTRY-296
> Project: Sentry
> Issue Type: Bug
> Affects Versions: 1.4.0
> Reporter: Lenni Kuff
> Assignee: Colin Ma
> Labels: grant/revoke
> Fix For: 1.6.0
>
> Attachments: SENTRY-296.001.patch, SENTRY-296.002.patch,
> SENTRY-296.003.patch, SENTRY-296.004.patch, SENTRY-296.005.patch,
> SENTRY-296.006.patch, SENTRY-296.007.patch, SENTRY-296.008.patch,
> SENTRY-296.009.patch, SENTRY-296.010.patch, SENTRY-296.011.patch,
> SENTRY-296.012.patch
>
>
> The Sentry Service Client does not allow for connection pooling because there
> is no way to:
> 1) Detect that a connection is broken
> 2) Re-open a broken connection
> This would allow for Sentry Service clients to implement connection pooling
> like:
> {code}
> if (!client.isOpen()) {
> client.reopen();
> }
> {code}
> Currently, the only way to ensure connections are not broken is to open a new
> connection before every RPC.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
