elasticsearch_ilm_action_steps.md

Phases+Actions

ILM phases+actions proceed sequentially. This implies if ILM gets stuck on a preceeding step it will never finish its sequence. A tabular format of the linked doc

actions\phase	hot	warm	cold	frozen	delete
set-priority	TRUE	TRUE	TRUE	f	f
unfollow	TRUE	TRUE	TRUE	TRUE	f
rollover	TRUE	f	f	f	f
read-only	TRUE	TRUE	TRUE	f	f
(searchable snapshot)	f	f	TRUE	f	f
allocate	f	TRUE	TRUE	f	f
migrate	f	TRUE	TRUE	f	f
shrink	TRUE	TRUE	f	f	f
force-merge	TRUE	TRUE	f	f	f
freeze	f	f	TRUE	f	f
searchable snapshots	TRUE	f	TRUE	TRUE	f
wait for snapshot	f	f	f	f	TRUE
delete	f	f	f	f	TRUE

Kindly note

• Searchable Snapshot shows twice because it sequentially orders differently uniquely for cold (and I don't know how to sohw that better).
• A common misconception which this table highlights: indices enter their target phase before performing the migrate action which means ILM is functioning as expected when shards are still on hot but index is in cold/unfollow.
• There are some backpressure sub-action settings, e.g. cold/searchable_snapshot may flag force_merge_index which performs that as if it had been flagged under hot/force_merge. This means the intended sequence would be on hot > force merge on hot > take searchable snapshot > allocate searchable snapshot onto cold. Note the index went into snapshot from hot and surfaced as searchable snapshot index on cold, but it did not go through the cold/migrate process directly.

Steps

👻 THIS IS ADVANCED. This is purposely not documented in the official Elasticsearch Docs by the Dev team. Proceed at your own risk.

These generally the progressive steps/action that are taken, but they are not the guaranteed order. Steps may invoke any other step causing backtracking/jumping.

action/ordered_steps	step
setPriority	set_priority
unfollow	branch-check-unfollow-prerequisites
...	wait-for-indexing-complete
...	wait-for-shard-tasks
...	pause-follower-index
...	close-follower-index
...	unfollow-follower-index
...	open-follower-index
...	wait-for-index-color
rollover	check-rollover-ready
...	attempt-rollover
...	wait-for-active-shards
...	update-rollover-lifecycle-data
...	set-indexing-complete
readonly	check-not-write-index
...	readonly
allocate	allocate
...	check-allocator
migrate	branch-check-skip-action
...	migrate
...	check-migration
shrink	check-not-write-index
...	wait-for-shard-history-leases
...	readonly
...	cleanup-shrink-index
...	generate-index-name
...	set-single-node-allocation
...	check-strink-allocation
...	shrink
...	shrunk-shards-allocated
...	copy-execution-state
...	isDataStreamBranchingStep
...	aliases
...	is-shrunken-index
...	replace-datastream-backing-index
...	delete
forceMerge	check-not-write-index
...	readonly
...	if codecChange == CodecService.BEST_COMPRESSION_CODEC - close-index - update-settings - open-index
...	wait-for-index-color
...	forcemerge
...	segment-count
freeze	check-not-write-index
...	freeze
searchablesnapshots	check-not-write-index
...	wait-for-shard-history-leases
...	if forceMergeIndex: - forcemerge - segment-count
...	generate-snapshot-name
...	cleanup-snapshot
...	create-snapshot
...	mount-snapshot
...	wait-for-index-color
...	copy-execution-state
...	copy-settings
...	isDataStreamBranchingStep
...	replace-datastream-backing-index
...	delete
...	swap-aliases
waitForSnapshot	wait-for-snapshot
delete	wait-for-shard-history-leases
...	delete