fix(java-client): fix client renew TGT thread leak when request table doesnt exist

[Samunroyu]

In the Kerberos environment, java client will check TGT and relogin
kerberos if needed every 10 seconds by a thread. But when client
requests table which doesnt exist will cause threads leak.

The root cause this problem is when Pegasus client close the client
but don't close TGT thread pool.

This patch adds KerberosProtocol close function.

[shalk]

Handle Exception is a good choice. Exception will make the schedule job not schedule again

But i am not quite understand the main cause about daemon
I don't kown the relatetionship between daemon thread with exception handler @Samunroyu

if you set daemon false, you should shutdown the threadpool when client jvm shutdown.

[Apache9]

Handle Exception is a good choice. Exception will make the schedule job not schedule again

But i am not quite understand the main cause about daemon I don't kown the relatetionship between daemon thread with exception handler @Samunroyu

if you set daemon false, you should shutdown the threadpool when client jvm shutdown.

I‘ve talked with @Samunroyu offline, the root cause here is we do not have a close method for KerberosProtocol class, so if we create the client many time due to kerberos authentication error, although we have already closed all the created client, the KerberosProtocol will not be closed, so the ExecutorService will be leaked and cause too many threads.

[shalk]

Handle Exception is a good choice. Exception will make the schedule job not schedule again
But i am not quite understand the main cause about daemon I don't kown the relatetionship between daemon thread with exception handler @Samunroyu
if you set daemon false, you should shutdown the threadpool when client jvm shutdown.

I‘ve talked with @Samunroyu offline, the root cause here is we do not have a close method for KerberosProtocol class, so if we create the client many time due to kerberos authentication error, although we have already closed all the created client, the KerberosProtocol will not be closed, so the ExecutorService will be leaked and cause too many threads.

It make sense. I get the case about thread leak.

if the pr finish and ci pass, click request , i will review again @Samunroyu

[Apache9]

Better introduce a unique suffix or prefix, otherwise if there are multiple client you will see multiple threads with the same name.

[Apache9]

Could still use lambda here.

    service.scheduleAtFixedRate(
        () -> {
            try {
              checkTGTAndRelogin();
            } catch (Exception e) {
              logger.warn(
                  "check TGT and ReLogin kerberos failed , will retry after {} seconds.",
                  CHECK_TGT_INTEVAL_SECONDS,
                  e);
            }
        }
        CHECK_TGT_INTEVAL_SECONDS,
        CHECK_TGT_INTEVAL_SECONDS,
        TimeUnit.SECONDS);

[shalk]

thread leak has fixed

[shalk]

from JDK8 , advise migration to the java.time classes.
String timestamp = LocalTime.now().format(DateTimeFormatter.ofPattern("HHmmss"));

[Samunroyu]

from JDK8 , advise migration to the java.time classes. String timestamp = LocalTime.now().format(DateTimeFormatter.ofPattern("HHmmss"));

ok, i have changed to use java.time.