Situation: Elasticsearch cluster crashed (thread pool and soon after heap memory depleted).
When investigating the heap dump it showed that the suggester was processing unusual queries like this:
+"\uffab-\uffaa/(+-\uffaa\uffab+(-\uffaa\uffab+(-\uffab\uffaa+(-+(\uffaa-\uffab"+(-\uffaa\uffab"+(+((-"\uffaa\uffab+(-\uffaa\uffab"+-(\uffaa\uffab/"*(-/\uffaa"+(\uffaa\uffab-"/-+(\uffaa-/\uffab"-(-\uffaa\uffab("+1
Now the questions are:
- Could it really be that a few of such queries lead to the described crash?
- If yes, why and how would it lead to a crash?
- Similar to question 2 but in more generality: what are the characteristics of queries that cause performance problems with the completion suggester?
- Additionally, how does the structure of the completion index relate to potential performance issues?
Any thoughts are appreciated