Addresses review feedback on the prior commit:
- The shared cache middleware should not hardcode `image/*`. Replace the
inline content-type check with a generic `cache.contentType` route
label (a prefix match). Each route that opts into caching now declares
`->label('cache.contentType', 'image/')` alongside `->label('cache', true)`.
- A public `Cache-Control: no-cache` bypass is a DDoS vector on expensive
pipelines (Imagick transforms, Chromium screenshots). Restrict the
bypass to privileged callers — API keys, admin console, and other
server-side contexts — so only trusted traffic can force-miss the cache.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A phpspy profile of a production databases worker showed the V20
backwards-compat request filter accounting for ~40% of in-request
samples on `databases.listDocuments` traffic. Two compounding causes:
1. `Request::getParams()` re-ran the entire filter chain on every
invocation. The framework and app call `getParams()` several times
per request (route param binding, `cacheIdentifier()`, action
injection, logging), so V20's recursive schema walk executed N times
with identical inputs.
2. Inside `V20::getRelatedCollectionKeys`, the `databases/$databaseId`
document was fetched at every recursion frame (up to
RELATION_MAX_DEPTH = 3), and sibling relationships pointing at the
same related collection each did their own `getDocument` call.
This commit:
- Memoizes the post-filter params on `Request`. The cache is
invalidated by `addFilter`, `resetFilters`, and `setRoute`. `Request`
is constructed per HTTP request (app/http.php), so the memo is
naturally request-scoped. Helps every request filter version, not
just V20.
- Splits V20's walk into an entry point that resolves the database
namespace once and a pure recursive helper.
- Caches the collection `attributes` array per
`(databaseNamespace, collectionId)` on the filter instance, so shared
related collections collapse to one `getDocument` call. Missing or
errored lookups are cached as `null` to avoid retry storms.
Redis stringifies scalars on save, so on a cache hit the `total` field
was served as a string. Flutter SDK (and any strictly-typed client) then
failed with `TypeError: "37": type 'String' is not a subtype of type 'int'`.
The cache-miss path returned an int from `count()`, which is why only
repeat requests with `ttl > 0` tripped the bug.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the per-call $measure closure with a single $dbStart
timestamp taken right before the fetch block and a single subtraction
right after it. Drops 6 lines of HOF indirection plus the $measure
variable, at the cost of including cache GET/SET time (~0.5–5ms) in
measurements when ttl > 0. For slow-query logging at a 100ms+
threshold that noise is negligible, and the default ttl=0 path has
no cache ops at all so the measurement is pure DB engine time.
The bracket captures the cursor lookup, find/count, and transaction
state calls — everything between "query parsed" and "fetch done",
as intended. processDocument's post-fetch relationship work is still
outside the bracket, matching the original design.
Moves the DB-duration measurement and afterQuery() hook from the
tablesDB-specific Rows/XList into the shared
Databases/Collections/Documents/XList base. Because TablesDB Rows and
DocumentsDB Documents both extend the legacy listDocuments base, a
single override now covers all three endpoints: legacy
listDocuments, listDocumentsDBDocuments, and tablesDB listRows.
TablesDB Rows drops the ~200-line action() duplicate and keeps only
the path/params/SDK overrides it needs, plus the extra
->inject('utopia') so its injection chain matches the new base action
signature. DocumentsDB Documents gets the same one-line inject
addition. Net -165 lines of duplication removed.
Behaviour is unchanged for CE (afterQuery() is a no-op); downstream
distributions overriding afterQuery() now observe every list-documents
/ list-rows call site for free.