Documentation
¶
Overview ¶
Package dryrun observes cache behaviour for a set of repositories without actually caching. Each observed request has its cache key recorded in a shadow store with the same sliding TTL the real cache uses, so metrics accurately report the hit / miss / coalesce rate the caching path would have produced. All requests are transparently proxied to upstream and no packfiles are written to disk. The dry-run allowlist reuses allowlist.Allowlist; the middleware's "enabled" check is `list.Bounded() && list.Contains(repo)`.
Index ¶
Constants ¶
This section is empty.
Variables ¶
var (
ErrFailedToCreateShadow = errors.New("failed to create dry-run shadow cache")
)
Functions ¶
This section is empty.
Types ¶
type Config ¶
type Config struct {
// MaxEntries caps the number of keys the shadow tracks. Well over the
// real cache's own MaxEntries so shadow evictions never fire before the
// real cache's would; that way "would-hit" isn't an under-count.
MaxEntries int64
// MaxIdleAge mirrors the real cache's sliding TTL.
MaxIdleAge time.Duration
// OnEviction, if set, receives "ttl" or "size" for each eviction.
OnEviction func(reason string)
}
Config configures a Shadow.
type Shadow ¶
type Shadow struct {
// contains filtered or unexported fields
}
Shadow is a byte-free stand-in for the real cache. It tracks whether a cache key would currently be a hit, applying the same sliding TTL policy the real cache uses so hit-rate metrics reflect real eviction behaviour. Values are empty; no packfiles are ever written.
func (*Shadow) Has ¶
Has reports whether the key is currently tracked. On a hit it also refreshes the entry's sliding TTL, matching the real cache's Get behaviour. The refresh is fire-and-forget: waiting for Ristretto's write buffer to drain would add latency to every hit on the request path, and eventual visibility of the refreshed TTL is fine because the next Has for the same key comes from a fresh HTTP request that takes far longer than the buffer flush.
Source Files