How to read from only one index and set the other one as write when searching an alias in ElasticSearch 7.6? - elasticsearch

Question

I know it's possible to define two indices in an alias where one index has the is_write_index set to true while the other has it set to false -
POST /_aliases
{
"actions" : [
{
"add" : {
"index" : "test_index_1",
"alias" : "my_alias",
"is_write_index": true
}
}
]
}
POST /_aliases
{
"actions" : [
{
"add" : {
"index" : "test_index_2",
"alias" : "my_alias",
"is_write_index": false
}
}
]
}
As you can see, I've defined two indices test_index_1 and test_index_2 where the first one is a write index while the second one isn't.
Now, I want to be able to query the my_alias in such a way that searches happen only on the test_index_2 which has the is_write_index set to false while I write data to test_index_1, instead of reading from both the indices, which is the default behaviour. Meaning, I don't wish the search results come from the index where is_write_index is set to true.
Is this possible? I've tried setting index.blocks.read to true on the write index, but then search queries on the alias fail with an exception. Instead, I wish reads on the alias query only from that index which has the is_write_index set to false.
How can I achieve this?

Answer 1

This can be achieved by using filtered aliases.
The way you do this is you apply a custom filter while adding the write index to the alias. The filter property defines the bool condition based on which data is filtered on this index and presented as a new view of the dataset in this index. All search queries on this index happen on this new view that Elastic creates. So, if you want to avoid reading from the index you're currently writing to, apply a filter that is never satisfied across any documents in your dataset or an exists filter on some dummy field.
POST /_aliases
{
"actions": [
{
"add": {
"index": "test_index_2",
"alias": "my_alias",
"is_write_index": true,
"filter": {
"bool": {
"must_not": {
"exists": {
"field": "<field_that_always_exists_in_your_documents>"
}
}
}
}
}
}
]
}
Once you're done writing the data, update the alias by removing the filter property to allow reads from both the indices.

Answer 2

You are using this feature in an incorrect fashion. If you use alias for search, it will always attempt to read across all underlying indices. is_write_index is provided as feature to support rollover and index patterns, where writes are happening to 1 index, but reads happen across all indices with same alias or index pattern.
If your intent is to load data into one index, but allow application to continue to read from old index, when data loading is going on, you should use 2 separate alias - one for read and one for write and device a strategy to swap alias pointing to the indices, after your data loading is completed.