Region counts have crept up (very slowly) to Hbase19 - 15 regions Hbase20 - 8 regions
So, would either continual reads or, (more likely) continual writes have such an effect on region splits and/or hbase balancing ? It is confusing as 3 new servers (NN, RS) were added to the cluster back in October 2020 and they were well-integrated within a day or so. -----Original Message----- From: Josh Elser <[email protected]> Sent: Sunday, January 24, 2021 10:22 PM To: [email protected] Subject: Re: Region server idle EXTERNAL Yes, each RegionServer has its own writeahead log (which is named with that RS' hostname). You'd want to look at the HBase master log, specifically for reasons as to why balancing is not naturally happening. It very well could be that you have other regions in transition (which may prevent balancing from happening). This is just one reason balancing may not naturally happening, but you should be able to see this in the active master log (potentially with enabling DEBUG on org.apache.hadoop.hbase, first). Don't forget about the hbase shell command to request the balancer to immediately run (so you can look for that logging at a specific point in time). On 1/18/21 7:23 AM, Marc Hoppins wrote: > I have been checking for days and there are no outstanding RITs. Region > servers do not have their own WAL files, do they? > > What gives me pause is that, although the affected servers (hbase19 & > hbase20) have 11 and 3 regions respectively, there must be very little > useable data as the requests per second are negligible for hbase19 and zero > for hbase20. > > I would have expected SOME movement to distribute data and work onto these > 'vacant' systems after more than 2 weeks. > > The circumstance behind hbase19 going offline is that a memory module had > vailed and dump data was constantly filling up tmp storage so the oncall guy > made the decision to shut the system down. Given that a lot of hbase work is > done in memory is there any possible way something still lingers in memory > somewhere that has not been flushed? > > As for hbase20, an IT guy decommissioned the host in the Cloudera console and > recommissioned it as a test to see if region balancing proceeded as normal. > Obviously, it hasn't. For obvious reasons, a second test has not been > performed. > > -----Original Message----- > From: Josh Elser <[email protected]> > Sent: Tuesday, January 12, 2021 4:56 PM > To: [email protected] > Subject: Re: Region server idle > > EXTERNAL > > Yes, in general, HDFS rebalancing will cause a decrease in the performance of > HBase as it removes the ability for HBase to short-circuit some read logic. > It should not, however, cause any kind of errors or lack of availability. > > You should feel free to investigate the RITs you have now, rather than wait > for a major compaction to finish. As a reminder, you can also force one to > happen now via the `major_compact` HBase shell command, for each table (or at > least the tables which are most important). Persistent RITs will prevent > balancing from happening, that may be your smoking gun. > > It may also be helpful for you to reach out to your vendor for support if you > have not done so already. > > On 1/12/21 6:11 AM, Marc Hoppins wrote: >> I read that HDFS balancing doesn't sit well with HBASE balancing. A >> colleague rebalanced HDFS on Friday. If I look for rebalance, despite me >> wandering through HBASE (Cloudera manager) it redirects to HDFS balance. >> >> I'd suggest I wait for major compaction to occur but who knows when that >> will be? Despite the default setting of 7 days in place, from what I read >> this will be dependent on no RITs being performed. As this is not just a >> working cluster but one of the more important ones, I am not sure if we can >> finish up any RITs to make the database 'passive' enough to perform a major >> compaction. >> >> Once again, experience in this area may be giv8ing me misinformation. >> >> -----Original Message----- >> From: Josh Elser <[email protected]> >> Sent: Monday, January 11, 2021 5:34 PM >> To: [email protected] >> Subject: Re: Region server idle >> >> EXTERNAL >> >> The Master stacktrace you have there does read as a bug, but it shouldn't be >> affecting balancing. >> >> That Chore is doing work to apply space quotas, but your quota here is only >> doing RPC (throttle) quotas. Might be something already fixed since the >> version you're on. I'll see if anything jumps out at me on Jira. >> >> If the Master isn't giving you any good logging, you could set the >> Log4j level to DEBUG for org.apache.hadoop.hbase (either via CM or >> the HBase UI for the active master, assuming that feature isn't >> disabled for security reasons in your org -- master.ui.readonly >> something something config property in hbase-site.xml) >> >> If DEBUG doesn't help, I'd set TRACE level for >> org.apache.hadoop.hbase.master.balancer. Granted, it might not be obvious to >> the untrained eye, but if you can share that DEBUG/TRACE after you manuall >> invoke the balancer again via hbase shell, it should be enough for those >> watching here. >> >> On 1/11/21 5:32 AM, Marc Hoppins wrote: >>> OK. So I tried again after running kinit and got the following: >>> >>> Took 0.0010 seconds >>> hbase(main):001:0> list_quotas >>> OWNER QUOTAS >>> USER => robot_urlrs TYPE => THROTTLE, >>> THROTTLE_TYPE => REQUEST_NUMBER, LIMIT => 100req/sec, SCOPE => MACHINE >>> 1 row(s) >>> >>> Not sure what to make of it but it doesn't seem like it is enough to >>> prevent balancing. There are other tables and (probably) other users. >>> >>> -----Original Message----- >>> From: Marc Hoppins <[email protected]> >>> Sent: Monday, January 11, 2021 9:52 AM >>> To: [email protected] >>> Subject: RE: Region server idle >>> >>> EXTERNAL >>> >>> I tried. Appears to have failed reading data from hbase:meta. These are >>> repeated errors for the whole run of list_quotas. >>> >>> A balance task was run on Friday. It took 9+ hours. The affected host had 6 >>> regions - no procedures/locks or processes were running for those 6 >>> regions. Today, that host has 8 regions. No real work being performed on >>> them. The other server - which went idle as a result of removing hbase19 >>> host from hbase and re-inserting to hbase - is still doing nothing and has >>> no regions assigned. >>> >>> I was su - hbase hbase shell to run it. >>> >>> **************** >>> >>> HBase Shell >>> Use "help" to get list of supported commands. >>> Use "exit" to quit this interactive shell. >>> For Reference, please visit: >>> http://hbase.apache.org/2.0/book.html#shell >>> Version 2.1.0-cdh6.3.2, rUnknown, Fri Nov 8 05:44:07 PST 2019 Took 0.0011 >>> seconds hbase(main):001:0> list_quotas >>> OWNER QUOTAS >>> org.apache.hadoop.hbase.client.RetriesExhaustedException: Failed after >>> attempts=8, exceptions: >>> Mon Jan 11 09:16:46 CET 2021, >>> RpcRetryingCaller{globalStartTime=1610353006298, pause=100, maxAttempts=8}, >>> javax.security.sasl.SaslException: Call to dr1-hbase18.jumb >>> >>> o.hq.com/10.1.140.36:16020 >>> failed on local exception: javax.security.sasl.SaslException: GSS initiate >>> failed [Caused by GSSException: No valid credentials provi >>> >>> ded (Mechanism level: Failed >>> to find any Kerberos tgt)] [Caused by javax.security.sasl.SaslException: >>> GSS initiate failed [Caused by GSSException: No valid credentia >>> >>> ls provided (Mechanism >>> level: Failed to find any Kerberos tgt)]] >>> Mon Jan 11 09:16:46 CET 2021, >>> RpcRetryingCaller{globalStartTime=1610353006298, pause=100, maxAttempts=8}, >>> java.io.IOException: Call to dr1-hbase18.jumbo.hq.com/ >>> >>> 10.1.140.36:16020 failed on >>> local exception: java.io.IOException: Can not send request because relogin >>> is in progress. >>> Mon Jan 11 09:16:46 CET 2021, >>> RpcRetryingCaller{globalStartTime=1610353006298, pause=100, maxAttempts=8}, >>> java.io.IOException: Call to dr1-hbase18.jumbo.hq.com/ >>> >>> 10.1.140.36:16020 failed on >>> local exception: java.io.IOException: Can not send request because relogin >>> is in progress. >>> Mon Jan 11 09:16:47 CET 2021, >>> RpcRetryingCaller{globalStartTime=1610353006298, pause=100, maxAttempts=8}, >>> java.io.IOException: Call to dr1-hbase18.jumbo.hq.com/ >>> >>> 10.1.140.36:16020 failed on >>> local exception: java.io.IOException: Can not send request because relogin >>> is in progress. >>> Mon Jan 11 09:16:47 CET 2021, >>> RpcRetryingCaller{globalStartTime=1610353006298, pause=100, maxAttempts=8}, >>> java.io.IOException: Call to dr1-hbase18.jumbo.hq.com/ >>> >>> 10.1.140.36:16020 failed on >>> local exception: java.io.IOException: Can not send request because relogin >>> is in progress. >>> Mon Jan 11 09:16:48 CET 2021, >>> RpcRetryingCaller{globalStartTime=1610353006298, pause=100, maxAttempts=8}, >>> javax.security.sasl.SaslException: Call to dr1-hbase18.jumb >>> >>> o.hq.com/10.1.140.36:16020 >>> failed on local exception: javax.security.sasl.SaslException: GSS initiate >>> failed [Caused by GSSException: No valid credentials provi >>> >>> ded (Mechanism level: Failed >>> to find any Kerberos tgt)] [Caused by javax.security.sasl.SaslException: >>> GSS initiate failed [Caused by GSSException: No valid credentia >>> >>> ls provided (Mechanism >>> level: Failed to find any Kerberos tgt)]] >>> Mon Jan 11 09:16:50 CET 2021, >>> RpcRetryingCaller{globalStartTime=1610353006298, pause=100, maxAttempts=8}, >>> java.io.IOException: Call to dr1-hbase18.jumbo.hq.com/ >>> >>> 10.1.140.36:16020 failed on >>> local exception: java.io.IOException: Can not send request because relogin >>> is in progress. >>> Mon Jan 11 09:16:54 CET 2021, >>> RpcRetryingCaller{globalStartTime=1610353006298, pause=100, maxAttempts=8}, >>> javax.security.sasl.SaslException: Call to dr1-hbase18.jumb >>> >>> o.hq.com/10.1.140.36:16020 >>> failed on local exception: javax.security.sasl.SaslException: GSS initiate >>> failed [Caused by GSSException: No valid credentials provi >>> >>> ded (Mechanism level: Failed >>> to find any Kerberos tgt)] [Caused by javax.security.sasl.SaslException: >>> GSS initiate failed [Caused by GSSException: No valid credentia >>> >>> ls provided (Mechanism >>> level: Failed to find any Kerberos tgt)]] >>> >>> -----Original Message----- >>> From: Stack <[email protected]> >>> Sent: Saturday, January 9, 2021 1:52 AM >>> To: Hbase-User <[email protected]> >>> Subject: Re: Region server idle >>> >>> EXTERNAL >>> >>> Looking at code around exception, can you check your quota settings? See >>> refguide on how to list quotas. Look for table or namespace that is empty >>> or non-existant and fill in missing portion. >>> >>> This is master-side log? It is from a periodic task so perhaps something >>> else is in the way of the non-assign? Anything else in there about >>> balancing or why we are skipping assign to these servers? Try a balance run >>> in the shell and then check master log to see why no work done? >>> >>> S >>> >>> On Fri, Jan 8, 2021 at 2:51 AM Marc Hoppins <[email protected]> wrote: >>> >>>> Apologies again. Here is the full error message. >>>> >>>> 2021-01-08 11:34:15,831 ERROR org.apache.hadoop.hbase.ScheduledChore: >>>> Caught error >>>> java.lang.IllegalStateException: Expected only one of namespace and >>>> tablename to be null >>>> at >>>> org.apache.hadoop.hbase.quotas.SnapshotQuotaObserverChore.getSnapshotsToComputeSize(SnapshotQuotaObserverChore.java:198) >>>> at >>>> org.apache.hadoop.hbase.quotas.SnapshotQuotaObserverChore._chore(SnapshotQuotaObserverChore.java:126) >>>> at >>>> org.apache.hadoop.hbase.quotas.SnapshotQuotaObserverChore.chore(SnapshotQuotaObserverChore.java:113) >>>> at >>>> org.apache.hadoop.hbase.ScheduledChore.run(ScheduledChore.java:186) >>>> at >>>> java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) >>>> at >>>> java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308) >>>> at >>>> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180) >>>> at >>>> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294) >>>> at >>>> org.apache.hadoop.hbase.JitterScheduledThreadPoolExecutorImpl$JitteredRunnableScheduledFuture.run(JitterScheduledThreadPoolExecutorImpl.java:111) >>>> at >>>> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) >>>> at >>>> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) >>>> at java.lang.Thread.run(Thread.java:748) >>>> >>>> -----Original Message----- >>>> From: Marc Hoppins <[email protected]> >>>> Sent: Friday, January 8, 2021 10:57 AM >>>> To: [email protected] >>>> Subject: RE: Region server idle >>>> >>>> EXTERNAL >>>> >>>> So, I tried decommission that RS and recommission it. No change. >>>> Server still idle. >>>> >>>> Tried decommission another server and see if HBASE sets itself right. >>>> Now I have two RS that are idle. >>>> >>>> ba-hbase18.jumbo.hq.com,16020,1604413480001 Tue Nov 03 15:24:40 CET >>>> 2020 1 s 2.1.0-cdh6.3.2 13 471 >>>> ba-hbase19.jumbo.hq.com,16020,1610095488001 Fri Jan 08 09:44:48 CET >>>> 2021 0 s 2.1.0-cdh6.3.2 0 6 >>>> ba-hbase20.jumbo.hq.com,16020,1610096850259 Fri Jan 08 10:07:30 CET >>>> 2021 0 s 2.1.0-cdh6.3.2 0 0 >>>> ba-hbase21.jumbo.hq.com,16020,1604414101652 Tue Nov 03 15:35:01 CET >>>> 2020 1 s 2.1.0-cdh6.3.2 15 447 >>>> >>>> From the logs: >>>> 2021-01-08 10:25:36,875 ERROR org.apache.hadoop.hbase.ScheduledChore: >>>> Caught error java.lang.IllegalStateException: Expected only one of >>>> namespace and tablename to be null >>>> >>>> This is reappearing in hbase master log >>>> >>>> M >>>> >>>> -----Original Message----- >>>> From: Sean Busbey <[email protected]> >>>> Sent: Thursday, January 7, 2021 7:30 PM >>>> To: Hbase-User <[email protected]> >>>> Subject: Re: Region server idle >>>> >>>> EXTERNAL >>>> >>>> Sounds like https://issues.apache.org/jira/browse/HBASE-24139 >>>> >>>> The description of that jira has a workaround. >>>> >>>> On Thu, Jan 7, 2021, 05:23 Marc Hoppins <[email protected]> wrote: >>>> >>>>> Hi all, >>>>> >>>>> I have a setup with 67 region servers. 29 Dec one system had to be >>>>> shut down to have EMM module swapped out - which took one work day. >>>>> Host was back online 30 Dec. >>>>> >>>>> My HBASE is very basic so I appreciate your patience. >>>>> >>>>> My understanding of the defaults that are setup is that a major >>>>> compaction should occur every 7 days. Moreover, do I assume that >>>>> more extensive balancing may occur after this happens? >>>>> >>>>> When I check (via hbase master UI) the status of HBASE, I see the >>>>> following: >>>>> >>>>> ServerName >>>>> >>>>> Start time >>>>> >>>>> Last contact >>>>> >>>>> Version >>>>> >>>>> Requests Per Second >>>>> >>>>> Num. Regions >>>>> >>>>> ba-hbase16.jumbo.hq.com,16020,1604413068640< >>>>> http://ba-hbase16.jumbo.hq.eset.com:16030/rs-status> >>>>> >>>>> Tue Nov 03 15:17:48 CET 2020 >>>>> >>>>> 3 s >>>>> >>>>> 2.1.0-cdh6.3.2 >>>>> >>>>> 46 >>>>> >>>>> 462 >>>>> >>>>> ba-hbase17.jumbo.hq.com,16020,1604413274393< >>>>> http://ba-hbase17.jumbo.hq.eset.com:16030/rs-status> >>>>> >>>>> Tue Nov 03 15:21:14 CET 2020 >>>>> >>>>> 1 s >>>>> >>>>> 2.1.0-cdh6.3.2 >>>>> >>>>> 19 >>>>> >>>>> 462 >>>>> >>>>> ba-hbase18.jumbo.hq.com,16020,1604413480001< >>>>> http://ba-hbase18.jumbo.hq.eset.com:16030/rs-status> >>>>> >>>>> Tue Nov 03 15:24:40 CET 2020 >>>>> >>>>> 2 s >>>>> >>>>> 2.1.0-cdh6.3.2 >>>>> >>>>> 62 >>>>> >>>>> 461 >>>>> >>>>> ba-hbase19.jumbo.hq.com,16020,1609326754985< >>>>> http://ba-hbase19.jumbo.hq.eset.com:16030/rs-status> >>>>> >>>>> Wed Dec 30 12:12:34 CET 2020 >>>>> >>>>> 2 s >>>>> >>>>> 2.1.0-cdh6.3.2 >>>>> >>>>> 0 >>>>> >>>>> 0 >>>>> >>>>> ba-hbase20.jumbo.hq.com,16020,1604413895967< >>>>> http://ba-hbase20.jumbo.hq.eset.com:16030/rs-status> >>>>> >>>>> Tue Nov 03 15:31:35 CET 2020 >>>>> >>>>> 2 s >>>>> >>>>> 2.1.0-cdh6.3.2 >>>>> >>>>> 62 >>>>> >>>>> 503 >>>>> >>>>> ba-hbase21.jumbo.hq.com,16020,1604414101652< >>>>> http://ba-hbase21.jumbo.hq.eset.com:16030/rs-status> >>>>> >>>>> Tue Nov 03 15:35:01 CET 2020 >>>>> >>>>> 3 s >>>>> >>>>> 2.1.0-cdh6.3.2 >>>>> >>>>> 59 >>>>> >>>>> 442 >>>>> >>>>> ba-hbase22.jumbo.hq.com,16020,1604414308289< >>>>> http://ba-hbase22.jumbo.hq.eset.com:16030/rs-status> >>>>> >>>>> Tue Nov 03 15:38:28 CET 2020 >>>>> >>>>> 0 s >>>>> >>>>> 2.1.0-cdh6.3.2 >>>>> >>>>> 40 >>>>> >>>>> 438 >>>>> >>>>> >>>>> Why, after more than 7 days, is this host not hosting more (any) regions? >>>>> Should I initiate some kind of rebalancing? >>>>> >>>>> Thanks in advance. >>>>> >>>>> M >>>>> >>>>
