Using ILM's move-to-step while ILM is stopped can still execute steps #81234
Description
When using ILM's move-to-step API, the API unconditionally executes certain types of steps when the index is moved into the step, for example:
PUT /_ilm/policy/fmpolicy
{
"policy": {
"phases": {
"hot": {
"actions": {
"rollover": {
"max_age": "60m"
}
}
},
"warm": {
"actions": {
"forcemerge": {
"max_num_segments": 1
}
}
},
"delete": {
"min_age": "1d",
"actions":{
"delete": {}
}
}
}
}
}
PUT /myindex-000001
{
"settings": {
"index.lifecycle.name": "fmpolicy",
"index.lifecycle.rollover_alias": "myalias"
},
"aliases": {
"myalias": {"is_write_index": true}
}
}
GET /myindex-*/_ilm/explain?human
POST /_ilm/stop
GET /_ilm/status
POST /_ilm/move/myindex-000001
{
"current_step": {
"phase": "hot",
"action": "rollover",
"name": "check-rollover-ready"
},
"next_step": {
"phase": "warm",
"action": "forcemerge",
"name": "readonly"
}
}
This should move from the rollover into the readonly step, however, when the move is executed, the readonly step is actually executed and the index ends up in:
{
"indices" : {
"myindex-000001" : {
"index" : "myindex-000001",
"managed" : true,
"policy" : "fmpolicy",
"phase" : "warm",
"action" : "forcemerge",
"step" : "segment-count",
...
}
}
}
It has advanced to the segment-count step rather than remaining in the readonly step (which is what I moved it to).
We should not execute steps when they are moved into if ILM is stopped, instead we should handle them correctly when ILM is restarted.