mintel/elasticsearch repository overview

⁠Mintel Kubernetes ( Mostly ) Elasticsearch Image

This repository contains the Mintel docker image that we use , mostly , within Kubernetes

⁠Build Details

The Image builds on-top of the Official Elasticsearch⁠ image and customize the following :

• It only run as unpriviliged user elasticsearch (uid:1000 / gid: 1000)
• The indexes are stored in /data if attaching a volume onto /data/ make sure the elasticsearch user can R/W to it
• JQ is added to the image
• elasticsearch-py python client library is added to the image
• run.sh entrypoint is customized to support a series of options / functionalities - see below⁠
• some management scripts are added to enable rolling restarts on Kubernetes - see below⁠ Those scripts are meant to be used as lifecycle hooks⁠ in kubernetes

⁠run.sh customizations

• On Graceful stop of the java process force sleep a few seconds to allow for other nodes to try to connect to the now stopped-pod and get a connection refuse. This is extremely important when stopping the active master - you can read more here⁠ Time to sleep can be customized by exporting POST_TERM_WAIT environment variable (Default to 15s)
• Install plugins at boot by specifying a comma separated list in ES_PLUGINS_INSTALL environment variable
• Support defining a custom SHARD_ALLOCATION_AWARENESS - TODO: To be improved soon with support with Kubernetes cloud zones
• Customization of Network DNS Caching TTL By default this java configuration from upstream will cache positive names forever and negative for a while ( TODO: how long? ) set NETWORK_ADDRESS_CACHE_TTL environment variable to define positive caching in seconds ( default to 3s ) set NETWORK_ADDRESS_CACHE_NEGATIVE_TTL environment variable to define negative caching in seconds ( default to 10s )

⁠lifecycle scripts

a Python script to manage various aspects of the Elasticsearch lifecycle is provided in /manage-es.py

A simple set of sh script to be used in kubernetes as lifecycle hooks⁠ is provided in

pre-stop hook

/stop-data-node.sh

python /manage-es.py pre-stop-data

post-start hook

/start-data-node.sh

python /manage-es.py persitent-settings
python /manage-es.py post-start-data

See the minikube⁠ example for a working example of the definition of those hooks

the following action are supported by the python script NOTE: In some cases running this scripts as lifecycle hook can lead to a cluster that can't startup

• For example all data action expect to be able to contact the cluster masters before proceeding. Is MASTER and DATA functions are on the same node this will never work

• pre-stop-data - Hook for Pre-Stop of a data node

    if mode.upper() == "ALLOCATION":
        # Sequence:
        # - Disable Shard Allocation
        # - Perform a Synced Flush
    elif mode.upper() == "DRAIN":
        # Sequence:
        # - set recovery settings
        # - drain Node
        # - Wait for 0 shards in relocating or initializing status

• post-start-data - Hook for Post-Start of a data node

    if mode.upper() == "ALLOCATION":                        
        # Sequence:                                                                                           
        # - wait for node to join cluster                                                                                                                                                    
        # - set recovery settings                           
        # - enable shards allocation                                                                          
        # - Wait for 0 Initializing or Relocating Shards ( Unassigned shards should be ok if this is cold startup of an elasticsearch cluster )                           
        # - remove temporary recovery settings  
    elif mode.upper() == "DRAIN":
        # Sequence:
        # - wait for node to join cluster
        # - set recovery settings
        # - undrain Node
        # - Wait for 0 Initializing or Relocating Shards ( Unassigned shards should be ok if this is cold startup of an elasticsearch cluster )                        
        # - remove temporary recovery settings

• pre-stop-master - Hook for Pre-Stop of a master node

Not implemented yet 

• post-start-master - Hook for Post-Start of a master node

Not implemented yet 

• peristent-settings - Set some elasticsearch persistent settings from a provided file

If path to a json persitent settings file is provided

push persistent settings to the cluster

⁠ENVironment Variables

⁠Startup Environment variables

• CLUSTER_NAME⁠
• NODE_NAME⁠
• NODE_MASTER⁠
• NODE_DATA⁠
• NETWORK_HOST⁠
• HTTP_CORS_ENABLE⁠
• HTTP_CORS_ALLOW_ORIGIN⁠
• NUMBER_OF_MASTERS⁠
• ES_GCLOG_FILE_COUNT - Number of GC log files to keep in the rotation.
• ES_GCLOG_FILE_PATH - Location of main GC file (e.g. data/gc.log).
• ES_GCLOG_FILE_SIZE - Max size of each rolled GC log (e.g. 64m).
• ES_JAVA_OPTS⁠
• ES_PLUGINS_INSTALL⁠ - comma separated list of Elasticsearch plugins to be installed. Example: ES_PLUGINS_INSTALL="repository-gcs,x-pack"
• SHARD_ALLOCATION_AWARENESS⁠
• SHARD_ALLOCATION_AWARENESS_ATTR⁠
• KUBERNETES_SHARD_ALLOCATION_AWARENESS - Enable SHARD Allocation awareness for a Kubernetes Cluster using labels
  • server = WORKER NODE NAME
    • zone = Kubernetes failure domain zone ( Only for cloud )
• MEMORY_LOCK⁠ - memory locking control - enable to prevent swap (default = true) .
• REPO_LOCATIONS⁠ - list of registered repository locations. For example "/backup" (default = []). The value of REPO_LOCATIONS is automatically wrapped within an [] and therefore should not be included in the variable declaration. To specify multiple repository locations simply specify a comma separated string for example "/backup", "/backup2".
• PROCESSORS⁠ - allow elasticsearch to optimize for the actual number of available cpus (must be an integer - default = 1)

⁠Elasticsearch settings environment variables

⁠Transient settings set during Maintenance

• WAIT_FOR_NODE_IN_CLUSTER - default: 180 - Seconds to wait for NODE to rejoin cluster
• WAIT_FOR_NO_SHARDS_RELOCATING - default: 1800 - Seconds to wait for cluster to have no Relocating nor Initializing shards
• NODE_CONCURRENT_INCOMING_RECOVERIES - default: ES Version defaut - set cluster.routing.allocation.node_concurrent_incoming_recoveries
• NODE_CONCURRENT_OUTGOING_RECOVERIES - default: ES Version defaut - set cluster.routing.allocation.node_concurrent_outgoing_recoveries
• NODE_INITIAL_PRIMARIES_RECOVERIES - default: ES Version defaut - set cluster.routing.allocation.node_initial_primaries_recoveries
• CLUSTER_CONCURRENT_REBALANCE - default: ES Version defaut - set cluster.routing.allocation.cluster_concurrent_rebalance

⁠Persistent Settings set by every node at startup

• PERSITENT_SETTINGS_FILE_PATH - path to a JSON file of persistent settings

⁠Other Startup Settings

• DISCOVERY_SERVICE - Elasticsearch discovery URL
• MAINTENANCE_MODE
  • None ( default ) - no Management of maintenance mode, pod will just be stopped by Kubernetes
  • Drain ( Drain local node ) - The node will be Drained ( moving all shards ) before proceeding with stop - NOTE: This need to finish before GracePeriod expire
  • Allocation ( Disable shard allocation ) - This will disable shards allocation as described https://www.elastic.co/guide/en/elasticsearch/reference/current/rolling-upgrades.html⁠

⁠run in minikube example

see example here⁠