Null Pointer Exception when searching by vector Id for an Id that doesn't exist

[bcrastnopol]

Describe the bug
I have a cosine-lsh mapped index, and when run nearest neighbor searches (exact and approximate) by vector id, I get the following error:

  "error" : {
    "root_cause" : [
      {
        "type" : "runtime_exception",
        "reason" : "Failed to retrieve vector at index [my-index] id [1] field [v]"
      }
    ],
    "type" : "runtime_exception",
    "reason" : "Failed to retrieve vector at index [my-index] id [1] field [v]",
    "caused_by" : {
      "type" : "null_pointer_exception",
      "reason" : "Cannot invoke \"java.util.Map.get(Object)\" because the return value of \"org.elasticsearch.action.get.GetResponse.getSourceAsMap()\" is null"
    }
  },
  "status" : 500
}

Expected behavior
Searching for a vector with an id that doesn't exist should not throw a 500 error

Environment (please complete the following information):

• Elastiknn version: 7.13.1.0 and 7.14.1.0
• OS: ubuntu linux

To Reproduce
Steps to reproduce the behavior:
1.Create an index

{
 "settings": {
   "index": {
     "number_of_shards": 1,          
     "elastiknn": true               
   }
 }
}

2. Add a mapping

{
   "properties": {
       "my_vec": {
           "type": "elastiknn_dense_float_vector",
           "elastiknn": {
               "dims": 100,                      
               "model": "lsh",                   
               "similarity": "cosine",             
               "L": 99,                            
               "k": 1                              
           }
       }
   }
}

3. Search for a vector that doesn't exist by Id

{
   "_source": [
     "vid"
 ], 
     "size": 10,
 "query": {
   
   "elastiknn_nearest_neighbors": {
     "model": "lsh",
     "similarity": "cosine",
     "candidates": 100,
     "field": "v",
     "vec": {
       "index": "my-index",
       "field": "v",
       "id": "1"
     }
   }
 }
}

4. See error

Additional context
I just want to say that this plugin is incredible! I'm running a very large cluster (100m+ vectors) and I've gotten great results compared to some some other nearest neighbor libraries. Plus, this has the added benefit of incremental index updates! Keep up the great work here!

[alexklibisz]

Thanks! It's probably a simple fix. I'll look into it in the next couple days.

[alexklibisz]

Hi @bcrastnopol , the issue should be resolved in the 7.14.1.1 release. I resolved in #311 and added some regression tests. Feel free to re-open this issue if it's still a problem.

I also opened #312 to look at some related latent issues and/or inconsistencies in exception handling.

Thanks for the kind words about the plugin. If you can, consider submitting a PR to add your use-case to the readme (here). I don't use the plugin in my day-to-day work so it's always neat to hear how it's being used.

[bcrastnopol]

Thank you for fixing this - it works great!

We're still evaluating a couple of solutions but I'm advocating for this. If we end up using it I'll find out what I'm allowed to share publicly and submit a PR!