HAWQ Filespaces and High Availability Enabled HDFS

Enabling the HDFS NameNode HA Feature
Using PXF with HDFS NameNode HA

If you initialized HAWQ without the HDFS High Availability (HA) feature, you can enable it by using the following procedure.

Enabling the HDFS NameNode HA Feature

To enable the HDFS NameNode HA feature for use with HAWQ, you need to perform the following tasks:

Enable high availability in your HDFS cluster.
Collect information about the target filespace.
Stop the HAWQ cluster and backup the catalog (Note: Ambari users must perform this manual step.)
Move the filespace location using the command line tool (Note: Ambari users must perform this manual step.)
Reconfigure ${GPHOME}/etc/hdfs-client.xml and ${GPHOME}/etc/hawq-site.xml files. Then, synchronize updated configuration files to all HAWQ nodes.
Start the HAWQ cluster and resynchronize the standby master after moving the filespace.

Step 1: Enable High Availability in Your HDFS Cluster

Enable high availability for NameNodes in your HDFS cluster. See the documentation for your Hadoop distribution for instructions on how to do this.

Note: If you’re using Ambari to manage your HDFS cluster, you can use the Enable NameNode HA Wizard. For example, this Hortonworks HDP procedure outlines how to do this in Ambari for HDP.

Step 2: Collect Information about the Target Filespace

A default filespace named dfs_system exists in the pg_filespace catalog and the parameter, pg_filespace_entry, contains detailed information for each filespace.

To move the filespace location to a HA-enabled HDFS location, you must move the data to a new path on your HA-enabled HDFS cluster.

Use the following SQL query to gather information about the filespace located on HDFS:

SELECT
    fsname, fsedbid, fselocation
FROM
    pg_filespace AS sp, pg_filespace_entry AS entry, pg_filesystem AS fs
WHERE
    sp.fsfsys = fs.oid AND fs.fsysname = 'hdfs' AND sp.oid = entry.fsefsoid
ORDER BY
    entry.fsedbid;

The sample output is as follows:

      fsname | fsedbid | fselocation
--------------+---------+-------------------------------------------------
cdbfast_fs_c | 0       | hdfs://hdfs-cluster/hawq//cdbfast_fs_c
cdbfast_fs_b | 0       | hdfs://hdfs-cluster/hawq//cdbfast_fs_b
cdbfast_fs_a | 0       | hdfs://hdfs-cluster/hawq//cdbfast_fs_a
dfs_system   | 0       | hdfs://test5:9000/hawq/hawq-1459499690
(4 rows)

The output contains the following:

HDFS paths that share the same prefix
Current filespace location

Note: If you see {replica=3} in the filespace location, ignore this part of the prefix. This is a known issue.

To enable HA HDFS, you need the filespace name and the common prefix of your HDFS paths. The filespace location is formatted like a URL.

If the previous filespace location is ‘hdfs://test5:9000/hawq/hawq-1459499690’ and the HA HDFS common prefix is 'hdfs://hdfs-cluster’, then the new filespace location should be 'hdfs://hdfs-cluster/hawq/hawq-1459499690’.
```
Filespace Name: dfs_system
Old location: hdfs://test5:9000/hawq/hawq-1459499690
New location: hdfs://hdfs-cluster/hawq/hawq-1459499690
```

Step 3: Stop the HAWQ Cluster and Back Up the Catalog

Note: Ambari users must perform this manual step.

When you enable HA HDFS, you are changing the HAWQ catalog and persistent tables. You cannot perform transactions while persistent tables are being updated. Therefore, before you move the filespace location, back up the catalog. This is to ensure that you do not lose data due to a hardware failure or during an operation (such as killing the HAWQ process).

If you defined a custom port for HAWQ master, export the PGPORT environment variable. For example:
```
export PGPORT=9000
```
Save the HAWQ master data directory, found in the hawq_master_directory property value from hawq-site.xml to an environment variable.
```
export MDATA_DIR=/path/to/hawq_master_directory
```
Disconnect all workload connections. Check the active connection with:
```
$ psql -p ${PGPORT} -c "SELECT * FROM pg_catalog.pg_stat_activity" -d template1
```
where ${PGPORT} corresponds to the port number you optionally customized for HAWQ master.

Issue a checkpoint:

$ psql -p ${PGPORT} -c "CHECKPOINT" -d template1

Shut down the HAWQ cluster:
```
$ hawq stop cluster -a -M fast
```
Copy the master data directory to a backup location:
```
$ cp -r ${MDATA_DIR} /catalog/backup/location
```
The master data directory contains the catalog. Fatal errors can occur due to hardware failure or if you fail to kill a HAWQ process before attempting a filespace location change. Make sure you back this directory up.

Step 4: Move the Filespace Location

Note: Ambari users must perform this manual step.

HAWQ provides the command line tool, hawq filespace, to move the location of the filespace.

If you defined a custom port for HAWQ master, export the PGPORT environment variable. For example:
```
export PGPORT=9000
```
Run the following command to move a filespace location:
```
$ hawq filespace --movefilespace default --location=hdfs://hdfs-cluster/hawq_new_filespace
```
Specify default as the value of the --movefilespace option. Replace hdfs://hdfs-cluster/hawq_new_filespace with the new filespace location.

Important: Potential Errors During Filespace Move

Non-fatal error can occur if you provide invalid input or if you have not stopped HAWQ before attempting a filespace location change. Check that you have followed the instructions from the beginning, or correct the input error before you re-run hawq filespace.

Fatal errors can occur due to hardware failure or if you fail to kill a HAWQ process before attempting a filespace location change. When a fatal error occurs, you will see the message, “PLEASE RESTORE MASTER DATA DIRECTORY” in the output. If this occurs, shut down the database and restore the ${MDATA_DIR} that you backed up in Step 4.

Step 5: Update HAWQ to Use NameNode HA by Reconfiguring hdfs-client.xml and hawq-site.xml

If you install and manage your cluster using command-line utilities, follow these steps to modify your HAWQ configuration to use the NameNode HA service.

Note: These steps are not required if you use Ambari to manage HDFS and HAWQ, because Ambari makes these changes automatically after you enable NameNode HA.

For command-line administrators:

Edit the ${GPHOME}/etc/hdfs-client.xml file on each segment and add the following NameNode properties:

<property>
 <name>dfs.ha.namenodes.hdpcluster</name>
 <value>nn1,nn2</value>
</property>

<property>
 <name>dfs.namenode.http-address.hdpcluster.nn1</name>
 <value>ip-address-1.mycompany.com:50070</value>
</property>

<property>
 <name>dfs.namenode.http-address.hdpcluster.nn2</name>
 <value>ip-address-2.mycompany.com:50070</value>
</property>

<property>
 <name>dfs.namenode.rpc-address.hdpcluster.nn1</name>
 <value>ip-address-1.mycompany.com:8020</value>
</property>

<property>
 <name>dfs.namenode.rpc-address.hdpcluster.nn2</name>
 <value>ip-address-2.mycompany.com:8020</value>
</property>

<property>
 <name>dfs.nameservices</name>
 <value>hdpcluster</value>
</property>

In the listing above:

Replace hdpcluster with the actual service ID that is configured in HDFS.
Replace ip-address-2.mycompany.com:50070 with the actual NameNode RPC host and port number that is configured in HDFS.
Replace ip-address-1.mycompany.com:8020 with the actual NameNode HTTP host and port number that is configured in HDFS.
The order of the NameNodes listed in dfs.ha.namenodes.hdpcluster is important for performance, especially when running secure HDFS. The first entry (nn1 in the example above) should correspond to the active NameNode.

Change the following parameter in the $GPHOME/etc/hawq-site.xml file:
```
<property>
    <name>hawq_dfs_url</name>
    <value>hdpcluster/hawq_default</value>
    <description>URL for accessing HDFS.</description>
</property>
```
In the listing above:
- Replace hdpcluster with the actual service ID that is configured in HDFS.
- Replace /hawq_default with the directory you want to use for storing data on HDFS. Make sure this directory exists and is writable.
Copy the updated configuration files to all nodes in the cluster (as listed in hawq_hosts).
```
$ hawq scp -f hawq_hosts hdfs-client.xml hawq-site.xml =:$GPHOME/etc/
```

Step 6: Restart the HAWQ Cluster and Resynchronize the Standby Master

Restart the HAWQ cluster:
```
$ hawq start cluster -a
```
Moving the filespace to a new location renders the standby master catalog invalid. To update the standby, resync the standby master. On the active master, run the following command to ensure that the standby master’s catalog is resynced with the active master.
```
$ hawq init standby -n -M fast
```

Using PXF with HDFS NameNode HA

If HDFS NameNode High Availability is enabled, use the HDFS Nameservice ID in the LOCATION clause <host> field when invoking any PXF CREATE EXTERNAL TABLE command. If the <port> is omitted from the LOCATION URI, PXF connects to the port number designated by the pxf_service_port server configuration parameter value (default is 51200).