Elasticsearch with Gluster Block Storage

In this blog we shall see

  1. Gluster block storage setup
  2. Elasticsearch Configuration with single node
  3. Testing
  4. Conclusion
  5. References

Before we begin,

  • In this post, I will try not to talk much about gluster block storage as that is not our main focus, one can look at my previous posts for more details on block storage terminology and architecture related information.
  • This post does not explain everything about Elasticsearch, it is just a POC that helps in setting up the gluster block storage as the backend persistent storage for Elasticsearch engine, and
  • Finally, be aware that gluster block storage is fresh and new and still in POC state.

All we need to perform this POC is 2 nodes with fedora 24 installed, and each having ~50G disk space.

Setup at a glance:

On Node1:
1. Install and run gluster and create a volume
2. Mount the volume created in step 1 and create a file of size 40G in the volume
3. Install and run tcmu-runner, create and export LUN using targetcli user:glfs handler
On Node2:
1. Discover and login to the target device exported in Node1
2. Notice the block device (/dev/sda) format it with xfs and mount
3. Install, configure Elasticsearch to use the mount point created in step 2 as data path and run it.
4. Play with the Elasticsearch engine by creating indices and querying.

Lets begin …

Gluster block storage setup

Installing glusterfs-server and configuring volume

Installing glusterfs 
# dnf install glusterfs-server
got glusterfs-server-3.8.5-1.fc24.x86_64.rpm

Run
# systemctl start glusterd
# systemctl status glusterd

Create a gluster volume
# gluster vol create block 10.70.42.151:/root/brick force
volume create: block: success: please start the volume to access data

Start the volume
# gluster vol start block
volume start: block: success

Mount the gluster volume
# mount.glusterfs localhost:/block /mnt/

Create a big file who play as target device
# fallocate -l 40G /mnt/elastic-media.img

# ls -l /mnt/
total 41943040
-rw-r--r--. 1 root root 42949672960 Nov 17 12:56 elastic-media.img

# df -Th
[...]
localhost:/block fuse.glusterfs 50G 41G 10G 81% /mnt

Tcmu-runner target emulation setup

Install tcmu-runner
# dnf install tcmu-runner

Run
# systemctl start tcmu-runner
# systemctl status tcmu-runner

Choose some iSCSI Qualified Name
# IQN=iqn.2016-11.org.gluster:10.70.42.151

Create the backend with glfs storage module
# targetcli /backstores/user:glfs create glfsLUN 40G block@10.70.42.151/elastic-media.img
Created user-backed storage object glfsLUN size 42949672960.

Create a target
# targetcli /iscsi create $IQN
Created target iqn.2016-11.org.gluster:10.70.42.151.
Created TPG 1.
Global pref auto_add_default_portal=true
Created default portal listening on all IPs (0.0.0.0), port 3260.

Share a glfs backed LUN without any auth checks
# targetcli /iscsi/$IQN/tpg1 set attribute generate_node_acls=1 demo_mode_write_protect=0
Parameter generate_node_acls is now '1'.
Parameter demo_mode_write_protect is now '0'.

Set/Export LUN
# targetcli /iscsi/$IQN/tpg1/luns create /backstores/user:glfs/glfsLUN
Created LUN 0.

# iptables -F

Initiator side setup (on Elasticsearch node) (NODE 2)

# dnf install iscsi-initiator-utils

Check existing block devices
# lsblk
NAME                       MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sr0                         11:0    1 1024M  0 rom  
vda                        252:0    0   40G  0 disk 
├─vda2                     252:2    0 39.5G  0 part 
│ ├─fedora_dhcp42--17-swap 253:1    0    4G  0 lvm  [SWAP]
│ └─fedora_dhcp42--17-root 253:0    0   15G  0 lvm  /
└─vda1                     252:1    0  500M  0 part /boot

Discovery and login to target
# iscsiadm -m discovery -t st -p 10.70.42.151 -l
10.70.42.151:3260,1 iqn.2016-06.org.gluster:10.70.42.151
Logging in to [iface: default, target: iqn.2016-06.org.gluster:10.70.42.151, portal: 10.70.42.151,3260] (multiple)
Login to [iface: default, target: iqn.2016-06.org.gluster:10.70.42.151, portal: 10.70.42.151,3260] successful.

Boom! got sda with 40G space 
# lsblk
NAME                       MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sr0                         11:0    1 1024M  0 rom  
sda                          8:0    0   40G  0 disk
vda                        252:0    0   40G  0 disk 
├─vda2                     252:2    0 39.5G  0 part 
│ ├─fedora_dhcp42--17-swap 253:1    0    4G  0 lvm  [SWAP]
│ └─fedora_dhcp42--17-root 253:0    0   15G  0 lvm  /
└─vda1                     252:1    0  500M  0 part /boot

Lets format the block device with xfs
#  mkfs.xfs /dev/sda

# mkdir /home/pkalever/block

Mount the block device
# mount /dev/sda /home/pkalever/block

# df -Th
Filesystem Type Size Used Avail Use% Mounted on
[...]
/dev/sda xfs 40G 0.2G 39.8G 1% /home/pkalever/block

Elasticsearch configuration with single node

Elasticsearch is an open-source, distributed, scalable, enterprise-grade search engine. Accessible through an extensive and elaborate API, Elasticsearch can power extremely fast searches that support your data discovery applications.

elasticsearch-2.3.4 (As it is compatible version with wiki dumps)

Download the rpm, this version is compatible with wiki indexes/dumps/docs
# wget https://download.elastic.co/elasticsearch/release/org/elasticsearch/distribution/rpm/elasticsearch/2.3.4/elasticsearch-2.3.4.rpm

Install Elasticsearch
# dnf install ./elasticsearch-2.3.4.rpm

Install Command-line JSON processor
# dnf install jq

Run
# sudo systemctl daemon-reload
# sudo systemctl enable elasticsearch.service
# sudo systemctl start elasticsearch.service

Check the status
# sudo systemctl status elasticsearch.service

Configure Elasticsearch to use gluster block mount directory for storage
Uncomment and edit the below parameters as per your choice
# sudo vi /etc/elasticsearch/elasticsearch.yml
cluster.name: gluster-block-17                 
node.name: node-17                             
path.data: /home/pkalever/block/data2     
path.logs: /home/pkalever/block/logs2

# mkdir  ~/block/data2  ~/block/log2

# /usr/share/elasticsearch/bin/plugin install analysis-icu

# sudo systemctl restart elasticsearch.service

Check the status
# sudo systemctl status elasticsearch.service

Testing

Simple test to make sure setup works

List the Indices
# curl -XGET http://localhost:9200/_cat/indices?v
health status index pri rep docs.count docs.deleted store.size pri.store.size 

Now let’s create an index name "bank"
# curl -XPUT http://localhost:9200/bank?pretty 
{
 "acknowledged" : true
}

# curl -XGET http://localhost:9200/_cat/indices?v
health status index pri rep docs.count docs.deleted store.size pri.store.size 
yellow open bank 5 1 0 0 650b 650b 

Note docs.count = 0 

Let’s now put something into our bank index.
In order to index a document, we must tell Elasticsearch which type in the index it should go to.
Let’s index a simple document into the bank index, "account" type, with an ID of 1 as follows:
# curl -XPUT http://localhost:9200/bank/account/1?pretty -d '
{
 "account_number": "999120999",
 "name": "pkalever"
}'

And the Response:
{
 "_index" : "bank",
 "_type" : "account",
 "_id" : "1",
 "_version" : 1,
 "_shards" : {
 "total" : 2,
 "successful" : 1,
 "failed" : 0
 },
 "created" : true
}

By looking at the response we can say that a new bank document was successfully created.
# curl -XGET http://localhost:9200/_cat/indices?v
health status index pri rep docs.count docs.deleted store.size pri.store.size 
yellow open bank 5 1 1 0 3.7kb 3.7kb

And now, Note docs.count = 1
 
Query a document
# curl -XGET http://localhost:9200/bank/account/1?pretty
{
 "_index" : "bank",
 "_type" : "account",
 "_id" : "1",
 "_version" : 1,
 "found" : true,
 "_source" : {
 "account_number" : "999120999",
 "name" : "pkalever"
 }
}

If we study the above commands carefully, we can actually see a pattern of how we access data in Elasticsearch.
That pattern can be summarized as follows:
<REST Verb> /<Index>/<Type>/<ID>

Delete the entry
# curl -XDELETE http://localhost:9200/bank/account/1?pretty

So we have manually created the indices and then added the documents, Lets now load some of the data sets/search index’s that Wikipedia provides.

Loading Wikipedia’s Search Index

In the very next script we do:
1. Delete if there is an index with name 'enwikiquote'
2. fetch the settings that en.wikiquote.org uses for its index and
   set them as template to create a new index
3. fetches the mapping for the content index and apply
# cat > run1.sh 
export es=localhost:9200
export site=en.wikiquote.org
export index=enwikiquote

curl -XDELETE $es/$index?pretty

curl -s 'https://'$site'/w/api.php?action=cirrus-settings-dump&format=json&formatversion=2' |
  jq '{
    analysis: .content.page.index.analysis,
    number_of_shards: 1,
    number_of_replicas: 0
  }' |
  curl -XPUT $es/$index?pretty -d @-

curl -s 'https://'$site'/w/api.php?action=cirrus-mapping-dump&format=json&formatversion=2' |
  jq .content |
  curl -XPUT $es/$index/_mapping/page?pretty -d @-

# ./run1.sh
{
  "acknowledged" : true
}
{
  "acknowledged" : true
}
{
  "acknowledged" : true
}

Now lets download the wiki dumps (the json formatted documents)
# wget https://dumps.wikimedia.org/other/cirrussearch/current/enwikiquote-20161114-cirrussearch-content.json.gz

Or you can go here and download whatever is needed for you https://dumps.wikimedia.org/other/cirrussearch/

In the very next script we
1. create a directory with name chunks and
2. extract 500 lines chunks from each file (250 lines metadata and 250 actual doc)
# cat > run2.sh 
export dump=enwikiquote-20161114-cirrussearch-content.json.gz
export index=enwikiquote

mkdir chunks
cd chunks
zcat ../$dump | split -a 10 -l 500 - $index


# ./run2.sh 
# ls chunks/
enwikiquoteaaaaaaaaaa  enwikiquoteaaaaaaaabd  enwikiquoteaaaaaaaacg  enwikiquoteaaaaaaaadj
enwikiquoteaaaaaaaaab  enwikiquoteaaaaaaaabe  enwikiquoteaaaaaaaach  enwikiquoteaaaaaaaadk
[...]
enwikiquoteaaaaaaaaba  enwikiquoteaaaaaaaacd  enwikiquoteaaaaaaaadg  enwikiquoteaaaaaaaaej
enwikiquoteaaaaaaaabb  enwikiquoteaaaaaaaace  enwikiquoteaaaaaaaadh  enwikiquoteaaaaaaaaek
enwikiquoteaaaaaaaabc  enwikiquoteaaaaaaaacf  enwikiquoteaaaaaaaadi

The loop in the script loads each file and deletes it after it's loaded. 
# cat > ./run3.sh
export es=localhost:9200
export index=enwikiquote
cd chunks
for file in *; do
  echo -n "${file}:  "
  took=$(curl -s -XPOST $es/$index/_bulk?pretty --data-binary @$file |
    grep took | cut -d':' -f 2 | cut -d',' -f 1)
  printf '%7s\n' $took
  [ "x$took" = "x" ] || rm $file
done

# ./run3.sh 
enwikiquoteaaaaaaaaaa:     9306
enwikiquoteaaaaaaaaab:    10607
enwikiquoteaaaaaaaaac:     6652
[...]
enwikiquoteaaaaaaaaaz:     4178
enwikiquoteaaaaaaaaba:     4800
enwikiquoteaaaaaaaabb:     4469
enwikiquoteaaaaaaaabc:     4349
[...]
enwikiquoteaaaaaaaabz:     8228
enwikiquoteaaaaaaaaca:     5152
enwikiquoteaaaaaaaacb:     4134
enwikiquoteaaaaaaaacc:     4510
[...]

List the indices 
# curl -XGET  http://localhost:9200/_cat/indices?v
health status index       pri rep docs.count docs.deleted store.size pri.store.size 
green  open   enwikiquote   1   0      28533            0      1.1gb          1.1gb

Query for page 1
# curl -XGET http://localhost:9200/enwikiquote/page/1?pretty

# curl -X GET  http://localhost:9200/enwikiquote/_search | less

Conclusion

This blog just showcases how Gluster block storage can be used as a backed persistent storage for Elasticsearch engine at POC level. More details will come by in further posts.

References

https://www.elastic.co/blog/loading-wikipedia

https://www.elastic.co/guide/en/elasticsearch/reference/current/_basic_concepts.html

https://www.elastic.co/guide/en/elasticsearch/reference/current/_create_an_index.html

Previous posts on gluster block storage

https://pkalever.wordpress.com/2016/06/23/gluster-solution-for-non-shared-persistent-storage-in-docker-container/

https://pkalever.wordpress.com/2016/06/29/non-shared-persistent-gluster-storage-with-kubernetes/

https://pkalever.wordpress.com/2016/08/16/read-write-once-persistent-storage-for-openshift-origin-using-gluster/

https://pkalever.wordpress.com/2016/11/04/gluster-as-block-storage-with-qemu-tcmu/

Advertisements