feat(cache): implement O(1) eviction policies and O(k) TTL cleanup #781
Conversation
This file contains hidden or 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
asaadbalum
requested review from
Xunzhuo,
rootfs and
wangchen615
as code owners
✅ Deploy Preview for vllm-semantic-router ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
👥 vLLM Semantic Team NotificationThe following members have been identified for the changed files in this PR and have been automatically assigned: 📁
|
Addresses Issue vllm-project#162: Implementation: - FIFO: Doubly-linked list queue for O(1) oldest entry selection - LRU: Doubly-linked list + hashmap for O(1) least-recently-used tracking - LFU: Frequency buckets with doubly-linked lists for O(1) least-frequently-used tracking - ExpirationHeap: Min-heap for O(k) TTL cleanup (k = expired entries) Performance (10k entries): - FIFO SelectVictim: ~4.3 ns/op - LRU SelectVictim: ~4.3 ns/op - LFU SelectVictim: ~6.2 ns/op Note: Issues 1 and 2 from vllm-project#162 were already addressed in PR vllm-project#347 and PR vllm-project#504. Signed-off-by: Asaad Balum <abalum@abalum-thinkpadp16vgen1.raanaii.csb>
asaadbalum
force-pushed
the
feature/cache-efficiency-improvements-162
branch
from
09d63b9 to
1a0a59d
Compare
Member
|
can you fix the CI? |
Signed-off-by: Asaad Balum <abalum@abalum-thinkpadp16vgen1.raanaii.csb>
Contributor
Author
@Xunzhuo |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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Outdated suggestions cannot be applied.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.
Summary
Implements O(1) eviction policies and O(k) TTL cleanup for the in-memory cache, addressing Issues 3 and 4 from #162.
Issues Addressed in This PR
Issue 3: Expensive TTL Cleanup ✅
Problem:
cleanupExpiredEntries()examined ALL cache entries - O(n) complexity.Solution: Implemented
ExpirationHeap(min-heap ordered by expiration time).Issue 4: Expensive Eviction ✅
Problem:
SelectVictim()scanned ALL entries to find eviction candidate - O(n) complexity.Solution: Implemented O(1) eviction policies using proper data structures:
Performance Improvement
Tests Added
TestFIFOPolicyOptimized- Tests FIFO O(1) operations (Evict, UpdateIndex, OnRemove)TestLRUPolicyOptimized- Tests LRU O(1) operations (Evict, UpdateIndex, OnRemove, OnAccess)TestLFUPolicyOptimized- Tests LFU O(1) operations (Evict, UpdateIndex, OnRemove, OnAccess)TestExpirationHeapOperations- Tests all heap operations (Size, PeekNext, UpdateExpiration, PopExpired)TestInMemoryCacheEviction- Tests eviction integration with InMemoryCacheIssues Already Addressed (Not in This PR)
Issue 1: Unnecessary Sorting in FindSimilar ✅
Problem:
sort.Slice(results, ...)added O(n log n) overhead to find best match.Already Fixed in PR #347: The sorting was removed. Now the best match is tracked during iteration - O(n) single pass, no sorting needed.
Issue 2: Linear Search for Similarity ✅
Problem: Computing dot product with every cache entry - O(n) per query.
Already Fixed in PR #504: HNSW (Hierarchical Navigable Small World) index was added for approximate nearest neighbor search - O(log n) average case.
FIX #162