docs: sync index-key definitions with the LoRA adapter partition#163
Open
EdHasNoLife wants to merge 1 commit into
Open
docs: sync index-key definitions with the LoRA adapter partition#163EdHasNoLife wants to merge 1 commit into
EdHasNoLife wants to merge 1 commit into
Conversation
The content/prefix key gained `adapter` in #159 — (tenant, model, hash_scheme, adapter, prefix_hash), and the chain/block key likewise — but several design/ops/reference docs still showed the pre-adapter 4-tuple. Update lookuproute-ranking, cachebackend-api, index-sizing, cachetenant-identity-and-quota, metrics, and reason-codes; add brief cardinality/quota notes where adapters multiply distinct keys. The (tenant, model, hash_scheme) serving SCOPE stays deliberately adapter-blind (TENANT_HOT / AFFINITY_HINT / diagnostics) — only the content key gains adapter. Docs only; no code change.
Codex reviewFiles reviewed
|
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Linear: CAC-245
What
#159 added
adapterto the index content key —(tenant, model, hash_scheme, adapter, prefix_hash), and the chain/block key likewise (""= base-model / non-LoRA). That PR's own touched docs (grpc-contract.md,kvevent-subscriber-wiring.md) describe it, but several design/ops/reference docs it did not touch still showed the pre-adapter 4-tuple. This syncs them.Changes (docs only)
docs/design/lookuproute-ranking.md— exact-key defs + the chain/block-hash key; a one-paragraph note thatadapterpartitions content (not the fingerprint) and that the serving scope stays adapter-blind.docs/design/cachebackend-api.md— the KV-event-subscriber section's index-key sentence.docs/operations/index-sizing.md— sizing/quota key + a LoRA note: adapters multiply distinct-key cardinality, so capacity guidance must account for the fan-out.docs/concepts/cachetenant-identity-and-quota.md— quota unit + entry key; note that identical content under N adapters = N distinct quota keys.docs/reference/metrics.md— theinferencecache_index_entriesdistinct-entry unit.docs/reference/reason-codes.md—PREFIX_MATCHcontent identity is now adapter-scoped;TENANT_HOT's exact-match key gainsadapterwhile its(tenant, model, hash_scheme)serving-scope check stays adapter-blind.What is deliberately NOT changed
The
(tenant, model, hash_scheme)serving scope —servingByScope/TENANT_HOT/AFFINITY_HINT/ theUNKNOWN_HASH_SCHEMEclassifier — stays adapter-blind by design (it answers replica-membership, not content residency; those surfaces shipmatched_tokens=0). Only the content/prefix key gainsadapter. No 3-tuple scope mention was touched.Follow-up to CAC-181 / #159. Code is unchanged and already correct — this is documentation consistency.
make verify-naming+verify-no-internal-refspass.