Merge branch '1.6.1-SNAPSHOT'
Conflicts:
docs/src/main/latex/accumulo_user_manual/chapters/troubleshooting.tex
Project: http://git-wip-us.apache.org/repos/asf/accumulo/repo
Commit: http://git-wip-us.apache.org/repos/asf/accumulo/commit/2556be27
Tree: http://git-wip-us.apache.org/repos/asf/accumulo/tree/2556be27
Diff: http://git-wip-us.apache.org/repos/asf/accumulo/diff/2556be27
Branch: refs/heads/master
Commit: 2556be27324d722483a0bf586711d6e6c86f47ad
Parents: 51742c6 366aef1
Author: Mike Drob <md...@cloudera.com>
Authored: Wed May 14 10:47:10 2014 -0400
Committer: Mike Drob <md...@cloudera.com>
Committed: Wed May 14 10:47:10 2014 -0400
----------------------------------------------------------------------
bin/generate_monitor_certificate.sh | 4 +-
.../org/apache/accumulo/core/conf/Property.java | 8 +-
.../main/asciidoc/chapters/troubleshooting.txt | 10 ++
.../accumulo/server/logger/LogFileKey.java | 25 +++
.../accumulo/server/logger/LogFileValue.java | 25 +++
.../accumulo/tserver/log/LocalWALRecovery.java | 176 +++++++++++++++++++
.../tserver/log/LocalWALRecoveryTest.java | 108 ++++++++++++
.../550e8400-e29b-41d4-a716-446655440000 | Bin 0 -> 91784 bytes
test/system/continuous/datanode-agitator.pl | 14 +-
9 files changed, 359 insertions(+), 11 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/accumulo/blob/2556be27/core/src/main/java/org/apache/accumulo/core/conf/Property.java
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/accumulo/blob/2556be27/docs/src/main/asciidoc/chapters/troubleshooting.txt
----------------------------------------------------------------------
diff --cc docs/src/main/asciidoc/chapters/troubleshooting.txt
index eec75fa,0000000..00c1106
mode 100644,000000..100644
--- a/docs/src/main/asciidoc/chapters/troubleshooting.txt
+++ b/docs/src/main/asciidoc/chapters/troubleshooting.txt
@@@ -1,758 -1,0 +1,768 @@@
+// Licensed to the Apache Software Foundation (ASF) under one or more
+// contributor license agreements. See the NOTICE file distributed with
+// this work for additional information regarding copyright ownership.
+// The ASF licenses this file to You under the Apache License, Version 2.0
+// (the "License"); you may not use this file except in compliance with
+// the License. You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+## Troubleshooting
+
+### Logs
+
+*Q*: The tablet server does not seem to be running!? What happened?
+
+Accumulo is a distributed system. It is supposed to run on remote
+equipment, across hundreds of computers. Each program that runs on
+these remote computers writes down events as they occur, into a local
+file. By default, this is defined in
++$ACCUMULO_HOME/conf/accumule-env.sh+ as +ACCUMULO_LOG_DIR+.
+
+*A*: Look in the +$ACCUMULO_LOG_DIR/tserver*.log+ file. Specifically, check
the end of the file.
+
+*Q*: The tablet server did not start and the debug log does not exists! What
happened?
+
+When the individual programs are started, the stdout and stderr output
+of these programs are stored in +.out+ and +.err+ files in
++$ACCUMULO_LOG_DIR+. Often, when there are missing configuration
+options, files or permissions, messages will be left in these files.
+
+*A*: Probably a start-up problem. Look in +$ACCUMULO_LOG_DIR/tserver*.err+
+
+### Monitor
+
+*Q*: Accumulo is not working, what's wrong?
+
+There's a small web server that collects information about all the
+components that make up a running Accumulo instance. It will highlight
+unusual or unexpected conditions.
+
+*A*: Point your browser to the monitor (typically the master host, on port
50095). Is anything red or yellow?
+
+*Q*: My browser is reporting connection refused, and I cannot get to the
monitor
+
+The monitor program's output is also written to .err and .out files in
+the +$ACCUMULO_LOG_DIR+. Look for problems in this file if the
++$ACCUMULO_LOG_DIR/monitor*.log+ file does not exist.
+
+*A*: The monitor program is probably not running. Check the log files for
errors.
+
+*Q*: My browser hangs trying to talk to the monitor.
+
+Your browser needs to be able to reach the monitor program. Often
+large clusters are firewalled, or use a VPN for internal
+communications. You can use SSH to proxy your browser to the cluster,
+or consult with your system administrator to gain access to the server
+from your browser.
+
+It is sometimes helpful to use a text-only browser to sanity-check the
+monitor while on the machine running the monitor:
+
+ $ links http://localhost:50095
+
+*A*: Verify that you are not firewalled from the monitor if it is running on
a remote host.
+
+*Q*: The monitor responds, but there are no numbers for tservers and tables.
The summary page says the master is down.
+
+The monitor program gathers all the details about the master and the
+tablet servers through the master. It will be mostly blank if the
+master is down.
+
+*A*: Check for a running master.
+
+### HDFS
+
+Accumulo reads and writes to the Hadoop Distributed File System.
+Accumulo needs this file system available at all times for normal operations.
+
+*Q*: Accumulo is having problems ``getting a block blk_1234567890123.'' How
do I fix it?
+
+This troubleshooting guide does not cover HDFS, but in general, you
+want to make sure that all the datanodes are running and an fsck check
+finds the file system clean:
+
+ $ hadoop fsck /accumulo
+
+You can use:
+
+ $ hadoop fsck /accumulo/path/to/corrupt/file -locations -blocks -files
+
+to locate the block references of individual corrupt files and use those
+references to search the name node and individual data node logs to determine
which
+servers those blocks have been assigned and then try to fix any underlying
file
+system issues on those nodes.
+
+On a larger cluster, you may need to increase the number of Xcievers for HDFS
DataNodes:
+
+[source,xml]
+<property>
+ <name>dfs.datanode.max.xcievers</name>
+ <value>4096</value>
+</property>
+
+*A*: Verify HDFS is healthy, check the datanode logs.
+
+### Zookeeper
+
+*Q*: +accumulo init+ is hanging. It says something about talking to
zookeeper.
+
+Zookeeper is also a distributed service. You will need to ensure that
+it is up. You can run the zookeeper command line tool to connect to
+any one of the zookeeper servers:
+
+ $ zkCli.sh -server zoohost
+ ...
+ [zk: zoohost:2181(CONNECTED) 0]
+
+It is important to see the word +CONNECTED+! If you only see
++CONNECTING+ you will need to diagnose zookeeper errors.
+
+*A*: Check to make sure that zookeeper is up, and that
++$ACCUMULO_HOME/conf/accumulo-site.xml+ has been pointed to
+your zookeeper server(s).
+
+*Q*: Zookeeper is running, but it does not say +CONNECTED+
+
+Zookeeper processes talk to each other to elect a leader. All updates
+go through the leader and propagate to a majority of all the other
+nodes. If a majority of the nodes cannot be reached, zookeeper will
+not allow updates. Zookeeper also limits the number connections to a
+server from any other single host. By default, this limit can be as small as
10
+and can be reached in some everything-on-one-machine test configurations.
+
+You can check the election status and connection status of clients by
+asking the zookeeper nodes for their status. You connect to zookeeper
+and ask it with the four-letter +stat+ command:
+
+----
+$ nc zoohost 2181
+stat
+Zookeeper version: 3.4.5-1392090, built on 09/30/2012 17:52 GMT
+Clients:
+ /127.0.0.1:58289[0](queued=0,recved=1,sent=0)
+ /127.0.0.1:60231[1](queued=0,recved=53910,sent=53915)
+
+Latency min/avg/max: 0/5/3008
+Received: 1561459
+Sent: 1561592
+Connections: 2
+Outstanding: 0
+Zxid: 0x621a3b
+Mode: standalone
+Node count: 22524
+----
+
+*A*: Check zookeeper status, verify that it has a quorum, and has not
exceeded maxClientCnxns.
+
+*Q*: My tablet server crashed! The logs say that it lost its zookeeper lock.
+
+Tablet servers reserve a lock in zookeeper to maintain their ownership
+over the tablets that have been assigned to them. Part of their
+responsibility for keeping the lock is to send zookeeper a keep-alive
+message periodically. If the tablet server fails to send a message in
+a timely fashion, zookeeper will remove the lock and notify the tablet
+server. If the tablet server does not receive a message from
+zookeeper, it will assume its lock has been lost, too. If a tablet
+server loses its lock, it kills itself: everything assumes it is dead
+already.
+
+*A*: Investigate why the tablet server did not send a timely message to
+zookeeper.
+
+#### Keeping the tablet server lock
+
+*Q*: My tablet server lost its lock. Why?
+
+The primary reason a tablet server loses its lock is that it has been pushed
into swap.
+
+A large java program (like the tablet server) may have a large portion
+of its memory image unused. The operation system will favor pushing
+this allocated, but unused memory into swap so that the memory can be
+re-used as a disk buffer. When the java virtual machine decides to
+access this memory, the OS will begin flushing disk buffers to return that
+memory to the VM. This can cause the entire process to block long
+enough for the zookeeper lock to be lost.
+
+*A*: Configure your system to reduce the kernel parameter _swappiness_ from
the default (60) to zero.
+
+*Q*: My tablet server lost its lock, and I have already set swappiness to
+zero. Why?
+
+Be careful not to over-subscribe memory. This can be easy to do if
+your accumulo processes run on the same nodes as hadoop's map-reduce
+framework. Remember to add up:
+
+* size of the JVM for the tablet server
+* size of the in-memory map, if using the native map implementation
+* size of the JVM for the data node
+* size of the JVM for the task tracker
+* size of the JVM times the maximum number of mappers and reducers
+* size of the kernel and any support processes
+
+If a 16G node can run 2 mappers and 2 reducers, and each can be 2G,
+then there is only 8G for the data node, tserver, task tracker and OS.
+
+*A*: Reduce the memory footprint of each component until it fits comfortably.
+
+*Q*: My tablet server lost its lock, swappiness is zero, and my node has lots
of unused memory!
+
+The JVM memory garbage collector may fall behind and cause a
+"stop-the-world" garbage collection. On a large memory virtual
+machine, this collection can take a long time. This happens more
+frequently when the JVM is getting low on free memory. Check the logs
+of the tablet server. You will see lines like this:
+
+ 2013-06-20 13:43:20,607 [tabletserver.TabletServer] DEBUG: gc
ParNew=0.00(+0.00) secs
+ ConcurrentMarkSweep=0.00(+0.00) secs
freemem=1,868,325,952(+1,868,325,952) totalmem=2,040,135,680
+
+When +freemem+ becomes small relative to the amount of memory
+needed, the JVM will spend more time finding free memory than
+performing work. This can cause long delays in sending keep-alive
+messages to zookeeper.
+
+*A*: Ensure the tablet server JVM is not running low on memory.
+
+### Tools
+
+The accumulo script can be used to run classes from the command line.
+This section shows how a few of the utilities work, but there are many
+more.
+
+There's a class that will examine an accumulo storage file and print
+out basic metadata.
+
+----
+$ ./bin/accumulo org.apache.accumulo.core.file.rfile.PrintInfo
/accumulo/tables/1/default_tablet/A000000n.rf
+2013-07-16 08:17:14,778 [util.NativeCodeLoader] INFO : Loaded the
native-hadoop library
+Locality group : <DEFAULT>
+ Start block : 0
+ Num blocks : 1
+ Index level 0 : 62 bytes 1 blocks
+ First key : 288be9ab4052fe9e
span:34078a86a723e5d3:3da450f02108ced5 [] 1373373521623 false
+ Last key : start:13fc375709e id:615f5ee2dd822d7a []
1373373821660 false
+ Num entries : 466
+ Column families : [waitForCommits, start, md major compactor 1,
md major compactor 2, md major compactor 3,
+ bringOnline, prep, md major compactor 4, md
major compactor 5, md root major compactor 3,
+ minorCompaction, wal, compactFiles, md root
major compactor 4, md root major compactor 1,
+ md root major compactor 2, compact, id,
client:update, span, update, commit, write,
+ majorCompaction]
+
+Meta block : BCFile.index
+ Raw size : 4 bytes
+ Compressed size : 12 bytes
+ Compression type : gz
+
+Meta block : RFile.index
+ Raw size : 780 bytes
+ Compressed size : 344 bytes
+ Compression type : gz
+----
+
+When trying to diagnose problems related to key size, the +PrintInfo+ tool
can provide a histogram of the individual key sizes:
+
+ $ ./bin/accumulo org.apache.accumulo.core.file.rfile.PrintInfo
--histogram /accumulo/tables/1/default_tablet/A000000n.rf
+ ...
+ Up to size count %-age
+ 10 : 222 28.23%
+ 100 : 244 71.77%
+ 1000 : 0 0.00%
+ 10000 : 0 0.00%
+ 100000 : 0 0.00%
+ 1000000 : 0 0.00%
+ 10000000 : 0 0.00%
+ 100000000 : 0 0.00%
+ 1000000000 : 0 0.00%
+ 10000000000 : 0 0.00%
+
+Likewise, +PrintInfo+ will dump the key-value pairs and show you the contents
of the RFile:
+
+ $ ./bin/accumulo org.apache.accumulo.core.file.rfile.PrintInfo --dump
/accumulo/tables/1/default_tablet/A000000n.rf
+ row columnFamily:columnQualifier [visibility] timestamp deleteFlag ->
Value
+ ...
+
+*Q*: Accumulo is not showing me any data!
+
+*A*: Do you have your auths set so that it matches your visibilities?
+
+*Q*: What are my visibilities?
+
+*A*: Use +PrintInfo+ on a representative file to get some idea of the
visibilities in the underlying data.
+
+Note that the use of +PrintInfo+ is an administrative tool and can only
+by used by someone who can access the underlying Accumulo data. It
+does not provide the normal access controls in Accumulo.
+
+If you would like to backup, or otherwise examine the contents of Zookeeper,
there are commands to dump and load to/from XML.
+
+ $ ./bin/accumulo org.apache.accumulo.server.util.DumpZookeeper --root
/accumulo >dump.xml
+ $ ./bin/accumulo org.apache.accumulo.server.util.RestoreZookeeper
--overwrite < dump.xml
+
+*Q*: How can I get the information in the monitor page for my cluster
monitoring system?
+
+*A*: Use GetMasterStats:
+
+ $ ./bin/accumulo org.apache.accumulo.test.GetMasterStats | grep Load
+ OS Load Average: 0.27
+
+*Q*: The monitor page is showing an offline tablet. How can I find out which
tablet it is?
+
+*A*: Use FindOfflineTablets:
+
+ $ ./bin/accumulo org.apache.accumulo.server.util.FindOfflineTablets
+ 2<<@(null,null,localhost:9997) is UNASSIGNED #walogs:2
+
+Here's what the output means:
+
++2<<+::
+ This is the tablet from (-inf, pass:[+]inf) for the
+ table with id 2. The command +tables -l+ in the shell will show table
ids for
+ tables.
+
++@(null, null, localhost:9997)+::
+ Location information. The
+ format is +@(assigned, hosted, last)+. In this case, the
+ tablet has not been assigned, is not hosted anywhere, and was once
+ hosted on localhost.
+
++#walogs:2+::
+ The number of write-ahead logs that this tablet requires for recovery.
+
+An unassigned tablet with write-ahead logs is probably waiting for
+logs to be sorted for efficient recovery.
+
+*Q*: How can I be sure that the metadata tables are up and consistent?
+
+*A*: +CheckForMetadataProblems+ will verify the start/end of
+every tablet matches, and the start and stop for the table is empty:
+
+ $ ./bin/accumulo org.apache.accumulo.server.util.CheckForMetadataProblems
-u root --password
+ Enter the connection password:
+ All is well for table !0
+ All is well for table 1
+
+*Q*: My hadoop cluster has lost a file due to a NameNode failure. How can I
remove the file?
+
+*A*: There's a utility that will check every file reference and ensure
+that the file exists in HDFS. Optionally, it will remove the
+reference:
+
+ $ ./bin/accumulo
org.apache.accumulo.server.util.RemoveEntriesForMissingFiles -u root --password
+ Enter the connection password:
+ 2013-07-16 13:10:57,293 [util.RemoveEntriesForMissingFiles] INFO : File
/accumulo/tables/2/default_tablet/F0000005.rf
+ is missing
+ 2013-07-16 13:10:57,296 [util.RemoveEntriesForMissingFiles] INFO : 1
files of 3 missing
+
+*Q*: I have many entries in zookeeper for old instances I no longer need.
How can I remove them?
+
+*A*: Use CleanZookeeper:
+
+ $ ./bin/accumulo org.apache.accumulo.server.util.CleanZookeeper
+
+This command will not delete the instance pointed to by the local
+conf/accumulo-site.xml+ file.
+
+*Q*: I need to decommission a node. How do I stop the tablet server on it?
+
+*A*: Use the admin command:
+
+ $ ./bin/accumulo admin stop hostname:9997
+ 2013-07-16 13:15:38,403 [util.Admin] INFO : Stopping server
12.34.56.78:9997
+
+*Q*: I cannot login to a tablet server host, and the tablet server will not
shut down. How can I kill the server?
+
+*A*: Sometimes you can kill a "stuck" tablet server by deleting its lock in
zookeeper:
+
+ $ ./bin/accumulo org.apache.accumulo.server.util.TabletServerLocks --list
+ 127.0.0.1:9997 TSERV_CLIENT=127.0.0.1:9997
+ $ ./bin/accumulo org.apache.accumulo.server.util.TabletServerLocks
-delete 127.0.0.1:9997
+ $ ./bin/accumulo org.apache.accumulo.server.util.TabletServerLocks -list
+ 127.0.0.1:9997 null
+
+You can find the master and instance id for any accumulo instances using the
same zookeeper instance:
+
+----
+$ ./bin/accumulo org.apache.accumulo.server.util.ListInstances
+INFO : Using ZooKeepers localhost:2181
+
+ Instance Name | Instance ID | Master
+---------------------+--------------------------------------+-------------------------------
+ "test" | 6140b72e-edd8-4126-b2f5-e74a8bbe323b |
127.0.0.1:9999
+----
+
+[[metadata]]
+### System Metadata Tables
+
+Accumulo tracks information about tables in metadata tables. The metadata for
+most tables is contained within the metadata table in the accumulo namespace,
+while metadata for that table is contained in the root table in the accumulo
+namespace. The root table is composed of a single tablet, which does not
+split, so it is also called the root tablet. Information about the root
+table, such as its location and write-ahead logs, are stored in ZooKeeper.
+
+Let's create a table and put some data into it:
+
+----
+shell> createtable test
+
+shell> tables -l
+accumulo.metadata => !0
+accumulo.root => +r
+test => 2
+trace => 1
+
+shell> insert a b c d
+
+shell> flush -w
+----
+
+Now let's take a look at the metadata for this table:
+
+ shell> table accumulo.metadata
+ shell> scan -b 3; -e 3<
+ 3< file:/default_tablet/F000009y.rf [] 186,1
+ 3< last:13fe86cd27101e5 [] 127.0.0.1:9997
+ 3< loc:13fe86cd27101e5 [] 127.0.0.1:9997
+ 3< log:127.0.0.1+9997/0cb7ce52-ac46-4bf7-ae1d-acdcfaa97995 []
127.0.0.1+9997/0cb7ce52-ac46-4bf7-ae1d-acdcfaa97995|6
+ 3< srv:dir [] /default_tablet
+ 3< srv:flush [] 1
+ 3< srv:lock [] tservers/127.0.0.1:9997/zlock-0000000001$13fe86cd27101e5
+ 3< srv:time [] M1373998392323
+ 3< ~tab:~pr [] \x00
+
+Let's decode this little session:
+
++scan -b 3; -e 3<+:: Every tablet gets its own row. Every row starts with
the table id followed by
+ +;+ or +<+, and followed by the end row split point for that tablet.
+
++file:/default_tablet/F000009y.rf [] 186,1+::
+ File entry for this tablet. This tablet contains a single file
reference. The
+ file is +/accumulo/tables/3/default_tablet/F000009y.rf+. It contains 1
+ key/value pair, and is 186 bytes long.
+
++last:13fe86cd27101e5 [] 127.0.0.1:9997+::
+ Last location for this tablet. It was last held on 127.0.0.1:9997, and
the
+ unique tablet server lock data was +13fe86cd27101e5+. The default balancer
+ will tend to put tablets back on their last location.
+
++loc:13fe86cd27101e5 [] 127.0.0.1:9997+::
+ The current location of this tablet.
+
++log:127.0.0.1+9997/0cb7ce52-ac46-4bf7-ae1d-acdcfaa97995 [] 127.0. ...+::
+ This tablet has a reference to a single write-ahead log. This file can be
found in
+ +/accumulo/wal/127.0.0.1+9997/0cb7ce52-ac46-4bf7-ae1d-acdcfaa97995+. The
value
+ of this entry could refer to multiple files. This tablet's data is
encoded as
+ +6+ within the log.
+
++srv:dir [] /default_tablet+::
+ Files written for this tablet will be placed into
+ +/accumulo/tables/3/default_tablet+.
+
++srv:flush [] 1+::
+ Flush id. This table has successfully completed the flush with the id of
+1+.
+
++srv:lock [] tservers/127.0.0.1:9997/zlock-0000000001\$13fe86cd27101e5+::
+ This is the lock information for the tablet holding the present lock.
This
+ information is checked against zookeeper whenever this is updated, which
+ prevents a metadata update from a tablet server that no longer holds its
+ lock.
+
++srv:time [] M1373998392323+::
+ This indicates the time time type (+M+ for milliseconds or +L+ for
logical) and the timestamp of the most recently written key in this tablet. It
is used to ensure automatically assigned key timestamps are strictly increasing
for the tablet, regardless of the tablet server's system time.
+
+`~tab:~pr [] \x00`::
+ The end-row marker for the previous tablet (prev-row). The first byte
+ indicates the presence of a prev-row. This tablet has the range (-inf,
+inf),
+ so it has no prev-row (or end row).
+
+Besides these columns, you may see:
+
++rowId future:zooKeeperID location+::
+ Tablet has been assigned to a tablet, but not yet loaded.
+
++~del:filename+::
+ When a tablet server is done use a file, it will create a delete marker
in the appropriate metadata table, unassociated with any tablet. The garbage
collector will remove the marker, and the file, when no other reference to the
file exists.
+
++~blip:txid+::
+ Bulk-Load In Progress marker.
+
++rowId loaded:filename+::
+ A file has been bulk-loaded into this tablet, however the bulk load has
not yet completed on other tablets, so this marker prevents the file from being
loaded multiple times.
+
++rowId !cloned+::
+ A marker that indicates that this tablet has been successfully cloned.
+
++rowId splitRatio:ratio+::
+ A marker that indicates a split is in progress, and the files are being
split at the given ratio.
+
++rowId chopped+::
+ A marker that indicates that the files in the tablet do not contain keys
outside the range of the tablet.
+
++rowId scan+::
+ A marker that prevents a file from being removed while there are still
active scans using it.
+
+### Simple System Recovery
+
+*Q*: One of my Accumulo processes died. How do I bring it back?
+
+The easiest way to bring all services online for an Accumulo instance is to
run the +start-all.sh+ script.
+
+ $ bin/start-all.sh
+
+This process will check the process listing, using +jps+ on each host before
attempting to restart a service on the given host.
+Typically, this check is sufficient except in the face of a hung/zombie
process. For large clusters, it may be
+undesirable to ssh to every node in the cluster to ensure that all hosts are
running the appropriate processes and +start-here.sh+ may be of use.
+
+ $ ssh host_with_dead_process
+ $ bin/start-here.sh
+
++start-here.sh+ should be invoked on the host which is missing a given
process. Like start-all.sh, it will start all
+necessary processes that are not currently running, but only on the current
host and not cluster-wide. Tools such as +pssh+ or
++pdsh+ can be used to automate this process.
+
++start-server.sh+ can also be used to start a process on a given host;
however, it is not generally recommended for
+users to issue this directly as the +start-all.sh+ and +start-here.sh+
scripts provide the same functionality with
+more automation and are less prone to user error.
+
+*A*: Use +start-all.sh+ or +start-here.sh+.
+
+*Q*: My process died again. Should I restart it via +cron+ or tools like
+supervisord+?
+
+*A*: A repeatedly dying Accumulo process is a sign of a larger problem.
Typically these problems are due to a
+misconfiguration of Accumulo or over-saturation of resources. Blind
automation of any service restart inside of Accumulo
+is generally an undesirable situation as it is indicative of a problem that
is being masked and ignored. Accumulo
+processes should be stable on the order of months and not require frequent
restart.
+
+
+### Advanced System Recovery
+
+#### HDFS Failure
+*Q*: I had disasterous HDFS failure. After bringing everything back up,
several tablets refuse to go online.
+
+Data written to tablets is written into memory before being written into
indexed files. In case the server
+is lost before the data is saved into a an indexed file, all data stored in
memory is first written into a
+write-ahead log (WAL). When a tablet is re-assigned to a new tablet server,
the write-ahead logs are read to
+recover any mutations that were in memory when the tablet was last hosted.
+
+If a write-ahead log cannot be read, then the tablet is not re-assigned. All
it takes is for one of
+the blocks in the write-ahead log to be missing. This is unlikely unless
multiple data nodes in HDFS have been
+lost.
+
+*A*: Get the WAL files online and healthy. Restore any data nodes that may
be down.
+
+*Q*: How do find out which tablets are offline?
+
+*A*: Use +accumulo admin checkTablets+
+
+ $ bin/accumulo admin checkTablets
+
+*Q*: I lost three data nodes, and I'm missing blocks in a WAL. I don't care
about data loss, how
+can I get those tablets online?
+
+See the discussion in <<metadata>>, which shows a typical metadata table
listing.
+The entries with a column family of +log+ are references to the WAL for that
tablet.
+If you know what WAL is bad, you can find all the references with a grep in
the shell:
+
+ shell> grep 0cb7ce52-ac46-4bf7-ae1d-acdcfaa97995
+ 3< log:127.0.0.1+9997/0cb7ce52-ac46-4bf7-ae1d-acdcfaa97995 []
127.0.0.1+9997/0cb7ce52-ac46-4bf7-ae1d-acdcfaa97995|6
+
+*A*: You can remove the WAL references in the metadata table.
+
+ shell> grant -u root Table.WRITE -t accumulo.metadata
+ shell> delete 3< log 127.0.0.1+9997/0cb7ce52-ac46-4bf7-ae1d-acdcfaa97995
+
+Note: the colon (+:+) is omitted when specifying the _row cf cq_ for the
delete command.
+
+The master will automatically discover the tablet no longer has a bad WAL
reference and will
+assign the tablet. You will need to remove the reference from all the
tablets to get them
+online.
+
+
+*Q*: The metadata (or root) table has references to a corrupt WAL.
+
+This is a much more serious state, since losing updates to the metadata table
will result
+in references to old files which may not exist, or lost references to new
files, resulting
+in tablets that cannot be read, or large amounts of data loss.
+
+The best hope is to restore the WAL by fixing HDFS data nodes and bringing
the data back online.
+If this is not possible, the best approach is to re-create the instance and
bulk import all files from
+the old instance into a new tables.
+
+A complete set of instructions for doing this is outside the scope of this
guide,
+but the basic approach is:
+
+* Use +tables -l+ in the shell to discover the table name to table id mapping
+* Stop all accumulo processes on all nodes
+* Move the accumulo directory in HDFS out of the way:
+ $ hadoop fs -mv /accumulo /corrupt
+* Re-initalize accumulo
+* Recreate tables, users and permissions
+* Import the directories under +/corrupt/tables/<id>+ into the new instance
+
+*Q*: One or more HDFS Files under /accumulo/tables are corrupt
+
+Accumulo maintains multiple references into the tablet files in the metadata
+tables and within the tablet server hosting the file, this makes it difficult
to
+reliably just remove those references.
+
+The directory structure in HDFS for tables will follow the general structure:
+
+ /accumulo
+ /accumulo/tables/
+ /accumulo/tables/!0
+ /accumulo/tables/!0/default_tablet/A000001.rf
+ /accumulo/tables/!0/t-00001/A000002.rf
+ /accumulo/tables/1
+ /accumulo/tables/1/default_tablet/A000003.rf
+ /accumulo/tables/1/t-00001/A000004.rf
+ /accumulo/tables/1/t-00001/A000005.rf
+ /accumulo/tables/2/default_tablet/A000006.rf
+ /accumulo/tables/2/t-00001/A000007.rf
+
+If files under +/accumulo/tables+ are corrupt, the best course of action is to
+recover those files in hdsf see the section on HDFS. Once these recovery
efforts
+have been exhausted, the next step depends on where the missing file(s) are
+located. Different actions are required when the bad files are in Accumulo
data
+table files or if they are metadata table files.
+
+*Data File Corruption*
+
+When an Accumulo data file is corrupt, the most reliable way to restore
Accumulo
+operations is to replace the missing file with an ``empty'' file so that
+references to the file in the METADATA table and within the tablet server
+hosting the file can be resolved by Accumulo. An empty file can be created
using
+the CreateEmpty utiity:
+
+ $ accumulo org.apache.accumulo.core.file.rfile.CreateEmpty
/path/to/empty/file/empty.rf
+
+The process is to delete the corrupt file and then move the empty file into
its
+place (The generated empty file can be copied and used multiple times if
necessary and does not need
+to be regenerated each time)
+
+ $ hadoop fs ârm /accumulo/tables/corrupt/file/thename.rf; \
+ hadoop fs -mv /path/to/empty/file/empty.rf
/accumulo/tables/corrupt/file/thename.rf
+
+*Metadata File Corruption*
+
+If the corrupt files are metadata files, see <<metadata>> (under the path
++/accumulo/tables/!0+) then you will need to rebuild
+the metadata table by initializing a new instance of Accumulo and then
importing
+all of the existing data into the new instance. This is the same procedure as
+recovering from a zookeeper failure (see <<zookeeper_failure>>), except that
+you will have the benefit of having the existing user and table authorizations
+that are maintained in zookeeper.
+
+You can use the DumpZookeeper utility to save this information for reference
+before creating the new instance. You will not be able to use
RestoreZookeeper
+because the table names and references are likely to be different between the
+original and the new instances, but it can serve as a reference.
+
+*A*: If the files cannot be recovered, replace corrupt data files with a empty
+rfiles to allow references in the metadata table and in the tablet servers to
be
+resolved. Rebuild the metadata table if the corrupt files are metadata files.
+
+[[zookeeper_failure]]
+#### ZooKeeper Failure
+*Q*: I lost my ZooKeeper quorum (hardware failure), but HDFS is still intact.
How can I recover my Accumulo instance?
+
+ZooKeeper, in addition to its lock-service capabilities, also serves to
bootstrap an Accumulo
+instance from some location in HDFS. It contains the pointers to the root
tablet in HDFS which
+is then used to load the Accumulo metadata tablets, which then loads all user
tables. ZooKeeper
+also stores all namespace and table configuration, the user database, the
mapping of table IDs to
+table names, and more across Accumulo restarts.
+
+Presently, the only way to recover such an instance is to initialize a new
instance and import all
+of the old data into the new instance. The easiest way to tackle this problem
is to first recreate
+the mapping of table ID to table name and then recreate each of those tables
in the new instance.
+Set any necessary configuration on the new tables and add some split points
to the tables to close
+the gap between how many splits the old table had and no splits.
+
+The directory structure in HDFS for tables will follow the general structure:
+
+ /accumulo
+ /accumulo/tables/
+ /accumulo/tables/1
+ /accumulo/tables/1/default_tablet/A000001.rf
+ /accumulo/tables/1/t-00001/A000002.rf
+ /accumulo/tables/1/t-00001/A000003.rf
+ /accumulo/tables/2/default_tablet/A000004.rf
+ /accumulo/tables/2/t-00001/A000005.rf
+
+For each table, make a new directory that you can move (or copy if you have
the HDFS space to do so)
+all of the rfiles for a given table into. For example, to process the table
with an ID of +1+, make a new directory,
+say +/new-table-1+ and then copy all files from +/accumulo/tables/1/*/*.rf+
into that directory. Additionally,
+make a directory, +/new-table-1-failures+, for any failures during the import
process. Then, issue the import
+command using the Accumulo shell into the new table, telling Accumulo to not
re-set the timestamp:
+
+ user@instance new_table> importdirectory /new-table-1
/new-table-1-failures false
+
+Any RFiles which were failed to be loaded will be placed in
+/new-table-1-failures+. Rfiles that were successfully
+imported will no longer exist in +/new-table-1+. For failures, move them back
to the import directory and retry
+the +importdirectory+ command.
+
+It is *extremely* important to note that this approach may introduce stale
data back into
+the tables. For a few reasons, RFiles may exist in the table directory which
are candidates for deletion but have
+not yet been deleted. Additionally, deleted data which was not compacted
away, but still exists in write-ahead logs if
+the original instance was somehow recoverable, will be re-introduced in the
new instance. Table splits and merges
+(which also include the deleteRows API call on TableOperations, are also
vulnerable to this problem. This process should
+*not* be used if these are unacceptable risks. It is possible to try to
re-create a view of the +accumulo.metadata+
+table to prune out files that are candidates for deletion, but this is a
difficult task that also may not be entirely accurate.
+
+Likewise, it is also possible that data loss may occur from write-ahead log
(WAL) files which existed on the old table but
+were not minor-compacted into an RFile. Again, it may be possible to
reconstruct the state of these WAL files to
+replay data not yet in an RFile; however, this is a difficult task and is not
implemented in any automated fashion.
+
+*A*: The +importdirectory+ shell command can be used to import RFiles from
the old instance into a newly created instance,
+but extreme care should go into the decision to do this as it may result in
reintroduction of stale data or the
+omission of new data.
+
++### Upgrade Issues
++
++*Q*: I upgraded from 1.4 to 1.5 to 1.6 but still have some WAL files on local
disk. Do I have any way to recover them?
++
++*A*: Yes, you can recover them by running the LocalWALRecovery utility on
each node that needs recovery performed. The utility
++will default to using the directory specified by +logger.dir.walog+ in your
configuration, or can be
++overriden by using the +--local-wal-directories+ option on the tool. It can
be invoked as follows:
++
++ $ACCUMULO_HOME/bin/accumulo
org.apache.accumulo.tserver.log.LocalWALRecovery
++
+### File Naming Conventions
+
+*Q*: Why are files named like they are? Why do some start with +C+ and others
with +F+?
+
+*A*: The file names give you a basic idea for the source of the file.
+
+The base of the filename is a base-36 unique number. All filenames in
accumulo are coordinated
+with a counter in zookeeper, so they are always unique, which is useful for
debugging.
+
+The leading letter gives you an idea of how the file was created:
+
++F+::
+ Flush: entries in memory were written to a file (Minor Compaction)
+
++M+::
+ Merging compaction: entries in memory were combined with the smallest
file to create one new file
+
++C+::
+ Several files, but not all files, were combined to produce this file
(Major Compaction)
+
++A+::
+ All files were compacted, delete entries were dropped
+
++I+::
+ Bulk import, complete, sorted index files. Always in a directory starting
with +b-+
+
+This simple file naming convention allows you to see the basic structure of
the files from just
+their filenames, and reason about what should be happening to them next, just
+by scanning their entries in the metadata tables.
+
+For example, if you see multiple files with +M+ prefixes, the tablet is, or
was, up against its
+maximum file limit, so it began merging memory updates with files to keep the
file count reasonable. This
+slows down ingest performance, so knowing there are many files like this
tells you that the system
+is struggling to keep up with ingest vs the compaction strategy which reduces
the number of files.
+
