perf(azure): batched SKU catalogue lookup eliminates N+1 in recommendation converters#81
Merged
cristim merged 1 commit intoApr 27, 2026
Conversation
…ation converters The four Azure recommendation converters (compute, database, cache, cosmosdb) previously left SKU-derived `Details` fields empty because populating them inline would have triggered an N+1 SDK call per recommendation. This change wires three of them to a lazily-cached, single-shot SKU catalogue lookup, gated by `sync.Once` and held for the client's lifetime. Per-service: - **Cache** (`services/cache`): caches `Properties.ShardCount` per SKU from `armredis.Client.NewListBySubscriptionPager`. Converter populates `CacheDetails.Shards` for Premium-tier clustered caches. - **Cosmos** (`services/cosmosdb`): caches the dominant Cosmos APIType (mongodb / cassandra / gremlin / table / sql) across the subscription's accounts via `armcosmos.DatabaseAccountsClient.NewListPager`, mapping `account.Kind` + `Capabilities`. Converter populates `NoSQLDetails.APIType`. Multi-API-type subscriptions leave APIType empty rather than guessing. - **Database** (`services/database`): caches engine version per SKU from `armsql.CapabilitiesClient.ListByLocation`, traversing ServerVersion → Editions → ServiceLevelObjectives. Converter populates `DatabaseDetails.EngineVersion`. Each converter calls a per-service `cachedSKULookup` (or `cachedAPIType` on cosmos). The catalogue is fetched ONCE per client lifetime; many converter calls in the same `GetRecommendations` run hit the in-memory map. A failed catalogue fetch leaves the cache nil and converters fall back to the previous empty-Details behaviour with a one-time WARN log — the conversion itself never fails, preserving the graceful-degradation contract. **Compute scoped back**: `common.ComputeDetails` (in `pkg/common/types.go`) exposes only InstanceType/Platform/Tenancy/Scope. The two enrichments the SKU catalogue would supply for compute are vCPU and MemoryGB — neither has a struct field to write into. Adding fields to the shared type would touch all 3 cloud providers' converters, the frontend, and the matchers, well outside this perf change's scope. Documented as a narrower follow-up in `known_issues/10_azure_provider.md`. Tests per service in the existing `*_test.go`: - `_PopulatesShardsFromSKUCache` / `_PopulatesAPIType` / `_PopulatesEngineVersion` — cache hit populates the previously-deferred field. - `_PagerErrorFallsBack` / `_CapabilitiesErrorFallsBack` — fetch error leaves the field empty and conversion still succeeds. - `_FetchedOnce` — many converter calls share a single catalogue fetch (the N+1 invariant). - The pre-existing `_PopulatesAllFields` tests now inject empty mock pagers so they don't hit the real Azure API on first lookup. Refs #49
|
Warning Rate limit exceeded
Your organization is not enrolled in usage-based pricing. Contact your admin to enable usage-based pricing to continue reviews beyond the rate limit, or try again in 31 minutes and 52 seconds. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (8)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Member
Author
|
@coderabbitai review |
✅ Actions performedReview triggered.
|
This was referenced Apr 25, 2026
Open
Member
Author
|
Two follow-up GitHub issues filed for the deferrals this PR explicitly scopes back:
Both reference back here and propose mirroring the cachedAPIType / cachedSKULookup patterns this PR establishes. |
cristim
added
triaged
cristim
added a commit
that referenced
this pull request
Apr 30, 2026
Adds optional VCPU (int) and MemoryGB (float64) fields to common.ComputeDetails so per-provider catalogue lookups can enrich compute recommendations with sizing data. Both fields use json omitempty — converters that don't yet wire a catalogue leave them at the zero value and the API payload stays clean. GetDetailDescription appends " (N vCPU / M GB)" when both values are > 0, otherwise returns the existing "platform/tenancy" form. Format uses %g so 16 GB renders as "16 GB" (not "16.000000 GB") while 0.5 GB SKUs render as "0.5 GB". Scope decision — schema-only foundation: - Azure: PR #81 (perf/azure-batched-sku-lookup) introduces the cachedSKULookup helper for cache/cosmos/database. Compute was scoped back from #81 because these fields didn't exist. With the schema in place, a follow-up PR can wire compute on top of #81 once it merges (filed as a separate issue post-merge to avoid an unmerged-PR dependency in this changeset). - AWS: ce.EC2InstanceDetails (Cost-Explorer) has no vCPU/Memory fields; populating would require a new ec2:DescribeInstanceTypes caller + cache + mocks. Substantial net-new code, tracked as a separate follow-up. - GCP: no compute recommendation converter exists today. N/A. - Frontend: api.Recommendation has no `details` field and the detail drawer renders a hard-coded list. Surfacing the new fields needs separate API + UI work; tracked as follow-up once at least one provider populates the fields. Tests: extend pkg/common/types_test.go::TestComputeDetails_GetDetail Description with 4 new table rows — VCPU-only, MemoryGB-only, integer-GB, and fractional-GB (Azure 0.5 GB SKU shape). Refs #82
This was referenced Apr 30, 2026
Open
cristim
added a commit
that referenced
this pull request
May 3, 2026
closes #148) (#229) PR #97 added VCPU + MemoryGB fields to common.ComputeDetails (with omitempty JSON tags). PR #81 introduced the lazy SKU-catalogue pattern for cache/cosmosdb/database but explicitly scoped back compute because the destination fields didn't exist. Both prerequisites are now in place; this wires Azure compute to the same pattern. Implementation mirrors cache/database verbatim for consistency: - New unexported vmSKUEntry{vCPUs, memoryGB}. - New skuCacheOnce sync.Once + skuCacheMap field on ComputeClient. - cachedSKULookup(ctx, skuName) lazily triggers the catalogue fetch on first call; subsequent calls are O(1) map lookups. ok=false on miss or fetch error. - fetchSKUCatalogue reuses the existing createResourceSKUsPager and isAvailableInRegion helpers (already used by GetValidResourceTypes), walks every page, and reduces virtualMachines SKUs into the map. Returns nil on error so the sync.Once-gated cache stays nil and converters fall back to the empty-fields path with a one-time WARN. - extractVMSKUCapabilities pulls vCPUs (Atoi) and MemoryGB (ParseFloat) from the SKU's Capabilities name/value list. Unparseable or missing capabilities → 0, treated as "unknown". - populateVMSKUMapFromPage was extracted out of fetchSKUCatalogue to stay under the project's gocyclo=10 threshold (matches the cache/database extraction pattern from PR #81). - convertAzureVMRecommendation now takes ctx (was _) and populates Details.VCPU/MemoryGB from the cache when both >0. The single caller GetRecommendations already passes ctx; no public API change. Behaviour preserved on failure: a transient ResourceSKUsClient error no longer breaks the conversion — VCPU/MemoryGB stay at 0 (omitempty hides them from API payloads), the rest of Details is populated from the recommendation payload as before, and a WARN is logged once per client lifetime. UX improvement: common.ComputeDetails.GetDetailDescription appends " (<vcpu> vCPU / <memory> GB)" when both fields are >0, so Azure VM recommendation summaries now include the size string instead of the SKU name alone. Tests added in providers/azure/services/compute/client_test.go (file-scoped vmSKUCatalogueMockPager keeps the shared mocks.MockResourceSKUsPager surface untouched, mirroring the cosmosdb / cache test convention): - TestComputeClient_ConvertAzureVMRecommendation_PopulatesVCPUAndMemoryFromSKUCache: catalogue hit populates VCPU=2, MemoryGB=8 for Standard_D2s_v3. - TestComputeClient_ConvertAzureVMRecommendation_PagerErrorFallsBack: fetch error leaves both fields at 0; conversion succeeds. - TestComputeClient_ConvertAzureVMRecommendation_NoMatchLeavesFieldsZero: catalogue miss leaves both fields at 0; conversion succeeds. - TestComputeClient_CachedSKULookup_FetchedOnce: 10 lookups trigger exactly 1 NextPage call (pins the N+1 invariant per PR #81). Out of scope (separate follow-ups, per the issue body): - AWS / GCP compute converter wiring. - Platform / Tenancy / Scope enrichment (require non-ResourceSKUs Azure data sources). Verification: gofmt clean; go vet ./... clean; gocyclo -over 10 finds no offenders in the modified file; go test github.com/LeanerCloud/CUDly/providers/azure/services/compute → 42 passing (4 new + all existing). The pre-existing providers/azure/services/search build failure (missing go.sum entry) is unrelated to this change. Closes #148
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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
Adds a lazy, single-shot SKU catalogue lookup to three of the four Azure recommendation converters (cache, cosmos, database) so the previously-deferred Details fields populate without N+1 SDK calls.
CacheDetails.Shardsfromarmredis.Properties.ShardCount.NoSQLDetails.APIType(sql / mongodb / cassandra / gremlin / table) from the dominant Cosmos accountKind+Capabilitiesin the subscription.DatabaseDetails.EngineVersionfromarmsql.CapabilitiesClient.ListByLocationtraversal.Each converter calls a per-service
cachedSKULookup(orcachedAPITypeon cosmos). The catalogue is fetched ONCE per client lifetime viasync.Once; many converter calls in the sameGetRecommendationsrun hit the in-memory map. A failed catalogue fetch leaves the cache nil and converters fall back to the previous empty-Details behaviour with a one-time WARN log — the conversion itself never fails.Compute scoped back:
common.ComputeDetails(inpkg/common/types.go) currently exposes only InstanceType / Platform / Tenancy / Scope. The two enrichments compute would supply (vCPU, MemoryGB) have no struct field to write into. Adding fields to a sharedpkg/commontype would touch all 3 cloud providers, the frontend, and the matchers — out of scope for this perf change. Tracked as a narrower follow-up inknown_issues/10_azure_provider.md.Test plan
New per-service tests in the existing
*_test.go:_PopulatesShardsFromSKUCache/_PopulatesAPIType(5 sub-cases) /_PopulatesEngineVersion— cache hit populates the previously-deferred field._PagerErrorFallsBack/_CapabilitiesErrorFallsBack— fetch error leaves the field empty and conversion still succeeds (graceful-degradation contract)._CachedSKULookup_FetchedOnce/_CachedAPIType_FetchedOnce— many converter calls share a single catalogue fetch (the N+1 invariant pinned in tests)._AmbiguousAPIType— multi-API-type Cosmos subscription leaves APIType empty rather than guessing._PopulatesAllFieldstests now inject empty mock pagers so they don't hit the real Azure API on first lookup.cd providers/azure && go test ./...— all green.go test ./...(root) — all green.go vet ./...— clean.Closes #49 for cache / cosmos / database. Compute follow-up filed as a narrower issue in the known-issues doc.