OpenSearch integration improvements #139
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
filipecosta90
changed the title
…ports it (supported for Lucene in OpenSearch version 2.13 and later)
… for anonymous_auth feature of opensearch)
… for anonymous_auth feature of opensearch)
LukasWestholt
approved these changes
Oct 2, 2024
| @@ -0,0 +1,87 @@ | |||
| [ | |||
| { | |||
| "name": "opensearch-default", | |||
There was a problem hiding this comment.
You need to rename the single-shard experiments from
"opensearch-" to "opensearch-single-shard-" so both are accessable.
There was a problem hiding this comment.
number_of_shards=1 is according to this already the default. Does those experiments then even do something different to opensearch-single-node-default-index.json?
There was a problem hiding this comment.
Should we maybe say number_of_replicas=0 for optimization? see https://repost.aws/knowledge-center/opensearch-indexing-performance
There was a problem hiding this comment.
Ah i see: number_of_replicas=0 is already set.
Comment on lines
+57
to
+58
| if "number_of_shards" in index_config: | ||
| index_settings["number_of_shards"] = 1 |
There was a problem hiding this comment.
"Tuples don't support item assignment"
I suggest:
index_settings = {
"knn": True,
"number_of_replicas": 0,
"refresh_interval": -1, # no refresh is required because we index all the data at once
}
index_config = collection_params.get("index")
# if we specify the number_of_shards on the config, enforce it. otherwise use the default
if "number_of_shards" in index_config:
index_settings["number_of_shards"] = 1
|
No real performace difference between old and new implementation. --- ../results/opensearch-old/output.json 2024-10-02 15:05:57.197185400 +0200
+++ ../results/opensearch-new/output.json 2024-10-02 15:52:09.001055200 +0200
@@ -1,59 +1,59 @@
[
{
"engine_name": "opensearch",
"setup_name": "opensearch-m-16-ef-128",
"dataset_name": "glove-100-angular",
- "upload_time": 505.1072022999997,
- "total_upload_time": 506.65230710000014,
+ "upload_time": 571.6861197999997,
+ "total_upload_time": 571.9920295000002,
"parallel": 1,
"engine_params": {
"knn.algo_param.ef_search": 128
},
- "mean_time": 0.08039978975999684,
- "mean_precisions": 0.819351,
- "std_time": 0.07328582483094778,
- "min_time": 0.021495099999810918,
- "max_time": 4.868069999999534,
- "rps": 12.39949293721894,
- "p95_time": 0.10517231999988325,
- "p99_time": 0.18999490199986213
+ "mean_time": 0.07982564825000299,
+ "mean_precisions": 0.8167789999999999,
+ "std_time": 0.07186505068791486,
+ "min_time": 0.008940600000642007,
+ "max_time": 6.208381400001599,
+ "rps": 12.489320049897197,
+ "p95_time": 0.10064215000093098,
+ "p99_time": 0.21390265599933658
},
{
"engine_name": "opensearch",
"setup_name": "opensearch-m-16-ef-128",
"dataset_name": "glove-100-angular",
- "upload_time": 505.1072022999997,
- "total_upload_time": 506.65230710000014,
+ "upload_time": 571.6861197999997,
+ "total_upload_time": 571.9920295000002,
"parallel": 1,
"engine_params": {
"knn.algo_param.ef_search": 256
},
- "mean_time": 0.08500446872999737,
- "mean_precisions": 0.816762,
- "std_time": 0.07344455333790761,
- "min_time": 0.026552799999990384,
- "max_time": 4.793960500000139,
- "rps": 11.71986972688328,
- "p95_time": 0.1365703799996481,
- "p99_time": 0.28884295299986684
+ "mean_time": 0.08107645404999203,
+ "mean_precisions": 0.814407,
+ "std_time": 0.020966374585830613,
+ "min_time": 0.028347600000415696,
+ "max_time": 1.7499939999997878,
+ "rps": 12.30329778171997,
+ "p95_time": 0.09037325000108466,
+ "p99_time": 0.09874180600041654
},
{
"engine_name": "opensearch",
"setup_name": "opensearch-m-16-ef-128",
"dataset_name": "glove-100-angular",
- "upload_time": 505.1072022999997,
- "total_upload_time": 506.65230710000014,
+ "upload_time": 571.6861197999997,
+ "total_upload_time": 571.9920295000002,
"parallel": 1,
"engine_params": {
"knn.algo_param.ef_search": 512
},
- "mean_time": 0.07950227081000166,
- "mean_precisions": 0.816762,
- "std_time": 0.027110040761532377,
- "min_time": 0.027317600000060338,
- "max_time": 1.4170608000003995,
- "rps": 12.539646869542393,
- "p95_time": 0.09963219500077684,
- "p99_time": 0.14407165300047384
+ "mean_time": 0.07992414966999531,
+ "mean_precisions": 0.814407,
+ "std_time": 0.010356757441751704,
+ "min_time": 0.025327800000013667,
+ "max_time": 0.42032800000015413,
+ "rps": 12.479517153224734,
+ "p95_time": 0.08913765000097555,
+ "p99_time": 0.09884917200071869
}
]
\ No newline at end of file |
Comment on lines
+81
to
+85
| # Update the index settings back to the default | ||
| refresh_interval = "1s" | ||
| cls.client.indices.put_settings( | ||
| index=OPENSEARCH_INDEX, | ||
| params={ | ||
| "timeout": 300, | ||
| }, | ||
| body={"index": {"refresh_interval": refresh_interval}}, |
There was a problem hiding this comment.
Is cls.client.indices.refresh(index=OPENSEARCH_INDEX) better?
There was a problem hiding this comment.
i believe it's best as is, meaning:
- we disable refresh during indexing
- we enable it back after it
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The bellow changes aim to: