Apache
/
hbase
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

HBASE-22625 documet use scan snapshot feature (#496) 

Fix feedback from Clay Baenziger.
Signed-off-by: Clay Baenziger <cwb@clayb.net>
This commit is contained in:
stack 2019-08-23 15:55:09 -07:00
parent 04feab97e0
commit 504fc52187
1 changed files with 20 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
src/main/asciidoc/_chapters/snapshot_scanner.adoc
View File

@ -31,7 +31,7 @@
In HBase, a scan of a table costs server-side HBase resources reading, formating, and returning data back to the client.
Luckily, HBase provides a TableSnapshotScanner and TableSnapshotInputFormat (introduced by link:https://issues.apache.org/jira/browse/HBASE-8369[HBASE-8369]),
which scan snapshot the HBase-written HFiles directly in the HDFS filesystem completely by-passing hbase. This access mode
which can scan HBase-written HFiles directly in the HDFS filesystem completely by-passing hbase. This access mode
performs better than going via HBase and can be used with an offline HBase with in-place or exported
snapshot HFiles.
@ -41,14 +41,14 @@ To read HFiles directly, the user must have sufficient permissions to access sna
TableSnapshotScanner provides a means for running a single client-side scan over snapshot files.
When using TableSnapshotScanner, we must specify a temporary directory to copy the snapshot files into.
The client user should have write permissions to this directory, and it should not be a subdirectory of
The client user should have write permissions to this directory, and the dir should not be a subdirectory of
the hbase.rootdir. The scanner deletes the contents of the directory once the scanner is closed.
.Use TableSnapshotScanner
====
[source,java]
----
Path restoreDir = new Path("XX"); // restore dir should not be a subdirectory HBase hbase.rootdir
Path restoreDir = new Path("XX"); // restore dir should not be a subdirectory of hbase.rootdir
Scan scan = new Scan();
try (TableSnapshotScanner scanner = new TableSnapshotScanner(conf, restoreDir, snapshotName, scan)) {
Result result = scanner.next();
@ -61,14 +61,14 @@ try (TableSnapshotScanner scanner = new TableSnapshotScanner(conf, restoreDir, s
====
=== TableSnapshotInputFormat
TableSnapshotInputFormat provide a way to scan over snapshot files in a MapReduce job.
TableSnapshotInputFormat provides a way to scan over snapshot HFiles in a MapReduce job.
.Use TableSnapshotInputFormat
====
[source,java]
----
Job job = new Job(conf);
Path restoreDir = new Path("XX"); // restore dir should not be a subdirectory HBase rootdir
Path restoreDir = new Path("XX"); // restore dir should not be a subdirectory of hbase.rootdir
Scan scan = new Scan();
TableMapReduceUtil.initTableSnapshotMapperJob(snapshotName, scan, MyTableMapper.class, MyMapKeyOutput.class, MyMapOutputValueWritable.class, job, true, restoreDir);
----
@ -77,31 +77,31 @@ TableMapReduceUtil.initTableSnapshotMapperJob(snapshotName, scan, MyTableMapper.
=== Permission to access snapshot and data files
Generally, only the HBase owner or the HDFS admin have the permission to access HFiles.
link:https://issues.apache.org/jira/browse/HBASE-18659[HBASE-18659] use HDFS ACLs to make HBase granted user have the permission to access the snapshot files.
link:https://issues.apache.org/jira/browse/HBASE-18659[HBASE-18659] uses HDFS ACLs to make HBase granted user have permission to access snapshot files.
==== link:https://hadoop.apache.org/docs/r2.7.1/hadoop-project-dist/hadoop-hdfs/HdfsPermissionsGuide.html#ACLs_Access_Control_Lists[HDFS ACLs]
HDFS ACLs supports an "access ACL", which defines the rules to enforce during permission checks, and a "default ACL",
which defines the ACL entries that new child files or sub-directories receive automatically during creation.
By HDFS ACLs, HBase sync granted users with read permission to HFiles.
Via HDFS ACLs, HBase syncs granted users with read permission to HFiles.
==== Basic idea
The HBase files are orginazed as the following ways:
The HBase files are organized in the following ways:
* {hbase-rootdir}/.tmp/data/{namespace}/{table}
* {hbase-rootdir}/data/{namespace}/{table}
* {hbase-rootdir}/archive/data/{namespace}/{table}
* {hbase-rootdir}/.hbase-snapshot/{snapshotName}
So the basic idea is to add or remove HDFS ACLs to files of
global/namespace/table directory when grant or revoke permission to global/namespace/table.
So the basic idea is to add or remove HDFS ACLs to files of the global/namespace/table directory
when grant or revoke permission to global/namespace/table.
See the design doc in link:https://issues.apache.org/jira/browse/HBASE-18659[HBASE-18659] for more details.
==== Configuration to use this feature
* Firstly, make sure that HDFS ACLs is enabled and umask is set to 027
* Firstly, make sure that HDFS ACLs are enabled and umask is set to 027
----
dfs.namenode.acls.enabled = true
fs.permissions.umask-mode = 027
@ -119,7 +119,7 @@ hbase.acl.sync.to.hdfs.enable=true
----
* Modify table scheme to enable this feature for a specified table, this config is
false by default for every table, this means the HBase granted acls will not be synced to HDFS
false by default for every table, this means the HBase granted ACLs will not be synced to HDFS
----
alter 't1', CONFIGURATION => {'hbase.acl.sync.to.hdfs.enable' => 'true'}
----
@ -137,11 +137,16 @@ HDFS has a config which limits the max ACL entries num for one directory or file
----
dfs.namenode.acls.max.entries = 32(default value)
----
The 32 entries include four fixed users for each directory or file: owner, group, other and mask. For a directory, the four users contain 8 ACL entries(access and default) and for a file, the four users contain 4 ACL entries(access). This means there are 24 ACL entries left for named users or groups.
The 32 entries include four fixed users for each directory or file: owner, group, other, and mask.
For a directory, the four users contain 8 ACL entries(access and default) and for a file, the four
users contain 4 ACL entries(access). This means there are 24 ACL entries left for named users or groups.
Based on this limitation, we can only sync up to 12 HBase granted users' ACLs. This means, if a table enable this feature, then the total users with table, namespace of this table, global READ permission should not be greater than 12.
Based on this limitation, we can only sync up to 12 HBase granted users' ACLs. This means, if a table
enables this feature, then the total users with table, namespace of this table, global READ permission
should not be greater than 12.
=====
=====
There are some cases that this coprocessor has not handled or could not handle, so the user HDFS ACLs are not syned normally. Such as a reference link to another hfile of other tables.
There are some cases that this coprocessor has not handled or could not handle, so the user HDFS ACLs
are not synced normally. It will not make a reference link to another hfile of other tables.
=====