clarification on describe index

[vihariazure]

below is my details of describe index
{
'm': '16',
'ef_construct': '100',
'index_type': 'HNSW',
'metric_type': 'L2',
'field_name': 'product_embeddings',
'index_name': 'milvus_index_uuid_trail_1',
'total_rows': 1000000,
'indexed_rows': 970000,
'pending_index_rows': 882500,
'state': 'Finished'
}
Details:
total_rows: 1000000

What it is: The total number of rows in the target field (product_embeddings).
Meaning: The field contains 1,000,000 vector entries in total.
indexed_rows: 970000

What it is: The number of rows that have been successfully indexed so far.
Meaning: Out of 1,000,000 rows, 970,000 have been indexed using the HNSW algorithm.
pending_index_rows: 882500

What it is: The number of rows still waiting to be indexed.
Meaning: A large number of rows are still pending indexing. This might indicate ongoing data ingestion or that the indexing process hasn’t been finalized for some rows.
state: 'Finished'

What it is: The state of the index-building process.
Meaning: 'Finished' typically indicates the index-building operation has been completed. However, the presence of pending_index_rows suggests that additional data is still awaiting indexing, possibly due to delayed ingestion or re-indexing processes.
Questions:

Pending Index Rows:

There are still pending_index_rows. This value generally be 0 after the index-building process is marked as 'Finished'?

Impact on Search Results:

If the state is 'Finished', does this mean I can search and retrieve results?
Since pending_index_rows is not 0, does this mean I will get results, but with reduced accuracy due to incomplete indexing?
Accuracy Considerations:

To achieve better accuracy, should I wait until pending_index_rows = 0 before performing searches?

Wait for Complete Indexing:

I used the following utility function to wait for the index-building process to complete until pending_index_rows = 0:
python
utility.wait_for_index_building_complete(collection_name=self.collection_name, index_name=self.index_name)

However, even after inserting data, the describe_index results still show a large number of pending_index_rows.
Parameter Optimization:

Do I need to adjust parameters like segment size or flush interval to ensure the index is ready as soon as data insertion is complete (so that pending_index_rows = 0)?

[xiaofan-luan]

If index is done, then you are ready to search.
Pending row could happen when you have new inserted data, or compaction happened.
For compaction segment, old indexed data will be there for serving so even if you have some pending rows that won't affect your performance