-
Retested with:
- Ansible version 2.9.12
- Target host OS: Ubuntu 16.04.6
-
DSE Installation is retested with the following DSE versions. Lower versions of DSE (5.0, 4.8, etc.) are already EOL (end-of-life) and EOSL (end-of-service-life).
- DSE 5.1.19
- DSE 6.0.12
- DSE 6.7.9
- DSE 6.8.2
-
DSE Upgrade is retested with the following DSE versions:
- DSE 6.0.12 -> DSE 6.7.9
- DSE 6.7.9 -> DSE 6.8.2
NOTE:
-
Upgrading DSE cluster is in general a fairly complicated process, especially when there are major version upgrade involved (e.g. DSE 5.1 -> DSE 6.0, etc.) and/or advanced workloads involved (e.g. Search, Analytics, Graph). What is demonstrated here is just for the simplest case and serves ONLY as an example.
-
This repository ONLY deals with DSE server installation/update and configuration. It does NOT handle any infrastructure/OS layer settings (including installing prerequisite software like Java). If you need to handle infrastructure/OS layer settings, please refer to my other repository terradse.
It is not a trivial task to provision a large DSE cluster, to manage the key configuration item changes (e.g. dse.yaml, cassandra.yaml, etc.), and to do version upgrade. Historically (pre DSE 5.0), there is really no out-of-the-box method to manage these tasks automatically. It is operationally a time-consuming and error-prone work to manually run these tasks on every DSE node instance, not even to mention that for many cases, you also have to consider the task execution sequence among different DES nodes (e.g. DC by DC, rack by rack, seed vs. no-seed and so on).
Starting from DSE 5.0, OpsCenter 6.0 (as part of DSE 5.0 release) introduces a new component called Life Cycle Manager (LCM) that is designed for efficient installation and configuration of a DSE cluster through a web based GUI. It aims to solve the above issue as an easy-to-use, out-of-the-box solution offered by DataStax. Although there are some limitations with the current version (OpsCenter 6.0.8 as the time of this writing), it is in general a good solution for automatic DSE cluster provisioning and configuration management. For more detailed information about this product, please reference to LCM's documentation page (https://docs.datastax.com/en/latest-opscenter/opsc/LCM/opscLCM.html).
Unfortunately, while the current version of LCM does a great job in provisioning new cluster and managing configuration changes, it does not support DSE version upgrade yet. Meanwhile, the way that LCM schedules the execution of a job (provisioning new node or making configuration change) across the entire DSE cluster is in serial mode. This means that before finishing the execution of a job on one DSE node, LCM won't kick off the job on another node. Although correct, such a scheduling mechanism is not very flexible and a little bit less efficient.
In an attempt to overcome the limitations of the (current version of) LCM and bring some flexibility in job execution scheduling across the entire DSE cluster, an Ansible framework (as presented in this document) is created for:
- Installing a brand new DSE cluster
- Upgrading an existing DSE cluster to a newer version
Note 1: The framework is currently based on Debian packaged DSE installation or upgrade.
Note 2: This framework does have some configuration management capabilities (e.g. to make changes in cassandra.yaml file), but it is not yet a general tool to manage all DSE components related configuration files. However, using Ansible makes this job much easier and quite straightforward.
Unlike some other orchestration and configuration management (CM) tools like Puppet or Chef, Ansible is very lightweight. It doesn't require any CM related modeuls, daemons, or agents to be installed on the managed machines. It is based on Python and pushes YAML based script commands over SSH protocol to remote hosts for execution. Ansible only needs to be installed on one workstation machine (could be your laptop) and as long as the SSH access from the workstation to the managed machines is enabled, Ansible based orchestration and CM will work.
Ansible uses an inventory file to group the host machines to be managed. Ansible commands (either through ad-hoc or in Ansible module) are executed against the systems or host machines specified in the inventory file. For more advanced orchestration and/or CM scenario, Ansible playbook is always used. By using the original words from Ansible documentation, the relationship among Ansible inventory, module, and playbook can be summarized as:
If Ansible modules are the tools in your workshop, playbooks are your instruction manuals, and your inventory of hosts are your raw material.
This framework is based on Ansible playbook. Following the Ansible best practice principles, the framework playbook directory structure is designed as below. Please note that this is just a starting point of this framework and the directory structure will be subject to future changes.
├── ansible.cfg
├── dse_install.yml
├── dse_upgrade.yml
├── group_vars
│ └── all
├── hosts
└── roles
├── datastax_pkg
│ ├── tasks
│ │ └── main.yml
│ └── vars
│ └── main.yml
├── dse_common
│ ├── tasks
│ │ └── main.yml
│ └── templates
│ └── seeds.j2
├── dse_instbin
│ └── tasks
│ └── main.yml
├── dse_updcfg
│ └── tasks
│ └── main.yml
├── dse_upgbin
│ └── tasks
│ └── main.yml
├── start_srvc
│ └── tasks
│ └── main.yaml
└── stop_srvc
└── tasks
└── main.yaml
The table below gives the high level description of the top level elements in the framework.
| Top Level Item Name | Type | Description |
|---|---|---|
| ansible.cfg | File | Configuration setting file to control Ansible execution behavior |
| dse_install.yml | File | Playbook for brand new DSE installation |
| dse_upgrade.yml | File | Playbook for upgrading DSE from an older version to a newer version |
| group_vars | Folder | Global variables that are available to all playbooks |
| hosts | File | Ansible inventory file |
| roles | Folder | Ansible roles (organization units) included in the playbooks |
The managed hosts can be put in different groups. Each managed host and the group it belongs to can have their own properties (or variables). Based on this understanding, the hosts file in this framework is organized such that:
- the managed hosts into DSE data-centers (DCs).
- each DC defines the following attributes:
- whether Solr workload isenabled in the DC
- whether Spark workload is enabled in the DC
- whether Graph workload is enabled in the DC (only applicable to DSE version 5.0+)
- whether the DC should be "auto_bootstrap"-ped (e.g. whehter the DC is to join an existing cluster)
- each managed host defines the following attributes:
- the private IP, if any, of the host
- whether the host is a seed node
- the DSE DC name that the host belongs to
- the rack name that the host belongs to
- whether the host is using VNode. if true (1), "intial_token" property (below) must be empty
- what is the initial_token value. if specified, the "vnode" property must be 0
Below is an exmple for a 2-DC DSE cluster. Eac DC has 2 nodes. The 1st DC (DC1) is VNode enabled and has only Cassandra workload. The 2nd DC (DC2) is using single-token setup and also has Solr workload enabled.
[dse:children]
dse_dc1
dse_dc2
[dse_dc1]
<node1_public_ip> private_ip=<node1_private_ip> seed=true dc=DC1 rack=RAC1 vnode=1 initial_token=
<node2_public_ip> private_ip=<node2_private_ip> seed=false dc=DC1 rack=RAC1 vnode=1 initial_token=
[dse_dc2]
<node3_public_ip> private_ip=<node3_private_ip> seed=true dc=DC2 rack=RAC1 vnode=0 initial_token=-9223372036854775808
<node4_public_ip> private_ip=<node4_private_ip> seed=false dc=DC2 rack=RAC1 vnode=0 initial_token=0
[dse_dc1:vars]
solr_enabled=0
spark_enabled=0
graph_enabled=0
auto_bootstrap=1
[dse_dc2:vars]
solr_enabled=1
spark_enabled=0
graph_enabled=0
auto_bootstrap=1
...
Other than the host-specific variables as defined in the inventory file (hosts). Group and global variables can also be put in group_vars sub-folder. In particular, group_vars/all file is used to specify the variables that are applicable to all managed hosts, no matter what groups the hosts belong to.
Within this framework, the following global variables are defined:
| Global Varialbe | Default Value | Description |
|---|---|---|
| backup_homedir | N/A | the home diretory for backing up DSE configuration files |
| dse_ver_major | N/A | the major DSE version (4.8, 5.0, 5.1, etc.) to be installed or upgraded to |
| dse_ver_minor | N/A | the minor DSE version (x.x.1, x.x.7, etc.) associated with the major version |
| dse_config_dir | /etc/dse | the home directory of DSE configuraion files |
| dse_default_dir | /etc/default | the home directory of DSE default setting file |
| dse_syslog_dir | /var/log/cassandra | the home directory of DSE system logs |
| vnode_token_num | 8 | num_token value setting for VNode setup |
| cluster_name | N/A | DSE cluster name |
| data_file_directories | N/A | Cassandra data directory (note: JBOD is NOT supported yet) |
| commitlog_directory | N/A | Cassandra commitlog directory |
| saved_caches_directory | N/A | Cassandra saved_caches directory |
| hints_directory | N/A | Cassandra hints directory (note: only applicable to DSE 5.0+) |
| cdc_raw_directory | N/A | Cassandra CDC raw directory (note: only applicable to DSE 5.0+) |
| metadata_directory | N/A | DSE metadata directory (note: only applicable to DSE 6.8+) |
Playbooks are the top-level, standalone execution units for ansbile. For this framework, there are only two playbooks. The playbook can be executed using the following command (from the root directory of this framework):
ansible-playbook -i ./hosts <playbook_name> --private-key=<ssh_private_key_file> -u <ssh_user_name>
-
dse_install.yml is used for installing a brand new DSE cluster at a version as defined in "dse_ver_target" global variable. At high level, what this playbook does is summarized as below:
- Calculate the seed list from the host-specific varaibles defined in the inventory file
- Install the specified DSE version (concurrently on all managed hosts)
- Configure DSE default workload, cassandra.yaml, and cassandra-rackdc.propertities files based on the global and host-specific varaibles
- Start seed node one by one. Verify the service is successfully up and running.
- Start non-seed node one by one. Verify the service is successfully up and running.
-
dse_upgrade.yml is used for upgrading an existing DSE cluster from the current version to a version as defined in "dse_ver_target" global variable. Please note that at the moment, the framework doesn't yet do sanity check around the DSE versions such as the target version must be newer than the current version. At high level, what this playbook does is summarized as below. Please note that due to the upgrade nature, the tasks are executed among all managed hosts in serial mode. This means that only when all steps are succesfully executed on one node then the execution moves to the next node.
- Stop DSE service and verify the service is successfully stopped
- Calculate the seed list from the host-specific varaibles defined in the inventory file
- Upgrade the DSE binary from the old version to the new version
- Configure DSE default workload, cassandra.yaml, and cassandra-rackdc.propertities files based on the global and host-specific varaibles
- Start DSE service and verify the service is successfully up and running.
The different steps in the playbooks are actually organized around (ansible) roles.
The Ansible roles defines the execution modules that can be included (and shared) in playbooks. The roles defined in this framework are:
| Role name | Description |
|---|---|
| datastax_pkg | DataStax APT package management |
| dse_common | Seed list calculation based on the inventory file definition for the managed hosts |
| dse_insbin | DSE Package binary installation |
| dse_upgbin | DSE version upgrade |
| dse_updcfg | Upgrade DSE key configuration files |
| start_srvc | Start DSE service and verification |
| stop_srvc | Stop DSE service and verification |
The scripts for these roles are quite straightforward. Some of the points worthy to mention are listed in the sub-sections below.
This role defines several variables related with DSE package management.
### Datastax repository file name
dse_repo_file: /etc/apt/sources.list.d/datastax.sources.list
The main goal of this role is to calculate the proper seed list for the DSE cluster based on the "seed" property information as defined for each managed host in the inventory file. The core calculation logic is defined in an ansible template file (templates/seeds.js2).
This role is shared for both installing brand new DSE cluster and version ugprading existing DSE cluster. Since the DataStax APT packages for version 4.x and 5.x are different (e.g. Graph component is new to DSE 5.0+), the script is actually designed to detect the DSE version, so the proper packages are installed.
This role is used for upgrading an existing DSE cluster to a newer version. It achieves so by doing the following 3 major steps of tasks:
-
Create a backup folder and backup the key DSE configuration files. If other configuration files need to be backed up, please modify this role (./roles/dse_upgbin/tasks/main.yml) accordingly.
- DSE default (e.g. /etc/default/dse)
- cassandra/*
- dse.yaml
- spark/*
-
Remove the old DSE APT package (dse-*)
-
Install the new DSE APT package (through role dse_instbin )
At the moment, this role does the following DSE configuration settings:
-
Set DSE default based on DSE DC workload property (as specified in the inventory file)
-
Set the following key "cassandra.yaml" settings based on the global and host specific variables
- cluster_name
- seeds
- endpoint_snitch (always set as GossipingPropertyFileSnitch)
- listen_address
- rpc_address
- commitlog_directory
- data_file_directories
- saved_cache_directory
- hints_directory (only applicable to DSE 5.0+)
- cdc_raw_directory (only applicable to DSE 5.0+)
- metadata_directory (only applicable to DSE 6.8+)
- num_tokens (if VNode setup is enabled)
- initial_token_value (if single-token setup is enabled)
- allocate_tokens_for_local_replication_factor (only applicable to DSE 5.0+)
-
Set the following key "cassandra-rackdc.properties" settings based on the host specific variables
- dc
- rack
Start DSE service; wait and verify it is completely up and running properly. The verification is done by checking Cassandra client port 9042 is in active LISTEN state.
Stop DSE service; wait and verify it is completely shutdown properly. The verification is done by checking that there is no DSE linux process is terminated.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent