[
https://issues.apache.org/jira/browse/HIVE-16873?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
BELUGA BEHR updated HIVE-16873:
-------------------------------
Description:
In {{org.apache.hadoop.hive.metastore.HiveMetaStore}} we have a {{Formatter}}
class (and its buffer) tied to every thread.
This {{Formatter}} is for logging purposes. I would suggest that we simply let
let the logging framework itself handle these kind of details and ditch the
buffer per thread.
{code}
public static final String AUDIT_FORMAT =
"ugi=%s\t" + // ugi
"ip=%s\t" + // remote IP
"cmd=%s\t"; // command
public static final Logger auditLog = LoggerFactory.getLogger(
HiveMetaStore.class.getName() + ".audit");
private static final ThreadLocal<Formatter> auditFormatter =
new ThreadLocal<Formatter>() {
@Override
protected Formatter initialValue() {
return new Formatter(new StringBuilder(AUDIT_FORMAT.length() * 4));
}
};
...
private static final void logAuditEvent(String cmd) {
final Formatter fmt = auditFormatter.get();
((StringBuilder) fmt.out()).setLength(0);
String address = getIPAddress();
if (address == null) {
address = "unknown-ip-addr";
}
auditLog.info(fmt.format(AUDIT_FORMAT, ugi.getUserName(),
address, cmd).toString());
}
{code}
was:
In {{org.apache.hadoop.hive.metastore.HiveMetaStore}} we have a {{Formatter}}
class tied to every thread for logging.
{code}
public static final String AUDIT_FORMAT =
"ugi=%s\t" + // ugi
"ip=%s\t" + // remote IP
"cmd=%s\t"; // command
public static final Logger auditLog = LoggerFactory.getLogger(
HiveMetaStore.class.getName() + ".audit");
private static final ThreadLocal<Formatter> auditFormatter =
new ThreadLocal<Formatter>() {
@Override
protected Formatter initialValue() {
return new Formatter(new StringBuilder(AUDIT_FORMAT.length() * 4));
}
};
...
private static final void logAuditEvent(String cmd) {
final Formatter fmt = auditFormatter.get();
((StringBuilder) fmt.out()).setLength(0);
String address = getIPAddress();
if (address == null) {
address = "unknown-ip-addr";
}
auditLog.info(fmt.format(AUDIT_FORMAT, ugi.getUserName(),
address, cmd).toString());
}
{code}
I would suggest that we simply let let the logging framework itself handle
these kind of details.
> Remove Thread Cache From Logging
> --------------------------------
>
> Key: HIVE-16873
> URL: https://issues.apache.org/jira/browse/HIVE-16873
> Project: Hive
> Issue Type: Improvement
> Components: Metastore
> Reporter: BELUGA BEHR
> Priority: Minor
> Attachments: HIVE-16873.1.patch
>
>
> In {{org.apache.hadoop.hive.metastore.HiveMetaStore}} we have a {{Formatter}}
> class (and its buffer) tied to every thread.
> This {{Formatter}} is for logging purposes. I would suggest that we simply
> let let the logging framework itself handle these kind of details and ditch
> the buffer per thread.
> {code}
> public static final String AUDIT_FORMAT =
> "ugi=%s\t" + // ugi
> "ip=%s\t" + // remote IP
> "cmd=%s\t"; // command
> public static final Logger auditLog = LoggerFactory.getLogger(
> HiveMetaStore.class.getName() + ".audit");
> private static final ThreadLocal<Formatter> auditFormatter =
> new ThreadLocal<Formatter>() {
> @Override
> protected Formatter initialValue() {
> return new Formatter(new StringBuilder(AUDIT_FORMAT.length() *
> 4));
> }
> };
> ...
> private static final void logAuditEvent(String cmd) {
> final Formatter fmt = auditFormatter.get();
> ((StringBuilder) fmt.out()).setLength(0);
> String address = getIPAddress();
> if (address == null) {
> address = "unknown-ip-addr";
> }
> auditLog.info(fmt.format(AUDIT_FORMAT, ugi.getUserName(),
> address, cmd).toString());
> }
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
