Skip Reasons

An item Houndarr evaluates but does not search usually gets an action=skipped row in the log with a reason string attached. Some reasons write at most one row per item per day, and season, artist, and author mode leave out a row another item of the same parent already accounted for; both are covered below. Most reasons are normal scheduling behavior, not errors.

Reasons

Reason string	Scope	What it means
on cooldown (Nd)	per-item	Missing item was searched less than Cooldown (days) ago.
on cutoff cooldown (Nd)	per-item	Cutoff item was searched less than Cutoff Cooldown ago.
on upgrade cooldown (Nd)	per-item	Upgrade item was searched less than Upgrade Cooldown (days) ago. Default 90 days.
not yet released	per-item	The release date is in the future. An item with no release date counts as released.
radarr reports not available	per-item	Radarr's own availability flag says the movie is not available yet.
radarr status indicates unreleased	per-item	Radarr's movie status is tba or announced and it is not flagged available.
whisparr v3 reports not available	per-item	Whisparr v3's availability flag says the scene is not available yet.
whisparr v3 status indicates unreleased	per-item	Whisparr v3's status is tba or announced and it is not flagged available.
future title not yet available	per-item	The Radarr or Whisparr v3 release year is still ahead, the status is not released, and it is not flagged available.
no series linked	per-item	Whisparr v2 returned an episode with no series attached, so it cannot be searched.
post-release grace (Nh)	per-item	Release date passed but the grace window (default 6 hours) has not elapsed.
waiting on post-release grace (Nh)	per-item	Season, artist, or author parent holding its early retry until a wanted item's grace window has certainly passed.
in hot retry window (Nh)	per-item	Missing item is inside its hot retry window, but the retry interval has not elapsed.
hourly limit reached (N/hr)	per-item	Missing pass hit Hourly Cap of N for the current hour.
cutoff hourly limit reached (N/hr)	per-item	Cutoff pass hit Cutoff Cap of N.
upgrade hourly limit reached (N/hr)	per-item	Upgrade pass hit Upgrade Cap of N.
tag filter (no included tag)	per-item	Tag Filter · Include is set and the item does not carry any matching tag.
tag filter (excluded tag)	per-item	Tag Filter · Exclude is set and the item carries one of those tags.
already in download queue	per-item	The *arr already has this item in its download queue (downloading, importing, or delayed).
queue backpressure (N/M)	cycle-level (info)	Download queue has N items, at or above Queue Limit of M. Entire cycle is skipped.
outside allowed time window	cycle-level (info)	Current time falls outside every window defined in Allowed Search Window. Entire cycle is skipped.

Cycle-level gates write one info row rather than a skipped one, and the supervisor sleeps until the next cycle; filter by Action = info to find them. Per-item gates write one row per candidate evaluated, except where the deduplication and context-mode rules below apply.

Release-aware retry

With the default hot retry window of 0, missing items skipped with not yet released or post-release grace (Nh) get one immediate retry on a later cycle once the release-timing gate clears, even when the normal missing cooldown has not fully elapsed. After that one retry, normal missing cooldown applies again.

When Hot Retry Window (hrs) is enabled, the latest post-release grace (Nh) row opens a short retry window. Houndarr can retry the item after Hot Retry Interval (hrs) elapses, still respecting batch size and the hourly cap. When the window closes, normal missing cooldown applies, except that an item the window never searched still takes its one retry.

Only dispatches and the release gate's own skips decide this, and a dispatch counts whether it succeeded or errored. A skip written by any other gate, such as hourly limit reached (N/hr), a cooldown row or a tag filter row, leaves the retry pending.

In season, artist, and author search mode these rows are logged under the parent, so the parent holds its early retry until every post-release grace (Nh) skip logged since its last search has certainly passed, and logs waiting on post-release grace (Nh) while it does. That keeps a just-aired episode from putting its whole season back in the search queue on every cycle while it waits out its own grace. Run Now searches anyway. Season 0 specials, and items whose parent the *arr did not report, are searched on their own id and never wait on a parent.

A not yet released row in these modes is handled differently. Sonarr, Whisparr v2, Lidarr and Readarr keep unreleased items out of their wanted lists, so one reaches Houndarr only while this host's clock trails the *arr's, and it says nothing about the parent. When another item of that parent passed the same release check on the cycle, the row is dropped rather than logged against the parent, whether or not that item went on to be searched. A parent with no item past the check on that cycle still logs the row, and still takes its early retry once one of them is released.

Each item in grace logs a row on every cycle that reaches it, so the wait ends about one grace window after the last such row. A parent whose items keep entering grace closer than about two grace windows apart can stop taking the early retry altogether and fall back to its ordinary cooldown; a daily show with Post-Release Grace (hrs) at 18 sits in that range. A cycle reaches any one item less often on a large library, which shortens the wait. While a wait is running, a Hot Retry Window (hrs) shorter than Post-Release Grace (hrs) closes before the wait ends, so set it longer if you want hot retries to land.

Cutoff and upgrade passes do not use this early retry. They always wait for their full cooldown.

Already in download queue

Right before sending a search, Houndarr checks the *arr's download queue and skips any item that already has an entry there: downloading, waiting to import, stuck on an import problem, waiting for a download client, or held back by a delay profile. Searching again mostly spends indexer hits, since the *arr turns down another grab for a queued item unless the quality profile allows upgrades and the search finds a better release, and a search sent by Houndarr would skip the delay profile entirely. No cooldown is recorded, so the item is searched again on a later cycle once its entry leaves the queue. An entry that never leaves, such as a stalled download, holds the item until you clear it in the *arr.

The queue is read at most once per cycle, and only when the cycle is about to search something. In season, artist, or author search mode, a queued item doesn't hold back the rest: the parent is skipped only when every one of its wanted items the cycle reaches is already queued. If the queue can't be read, the cycle searches as usual, writes a warning to the container log, and logs a download queue check (fetch failed) info row. A reverse proxy rule or an ACL can block that one endpoint while every other request keeps working, so the check stays off cycle after cycle and the row is what shows it on the Logs page. The row is throttled to one per instance every six hours, and the window is held in memory, so a restart starts it over.

Queue backpressure

Setting Queue Limit to a value above zero makes Houndarr check the download queue before each cycle. When the queue count meets or exceeds the limit, the cycle writes one queue backpressure (N/M) skip and sleeps. If the queue endpoint is unreachable, the cycle proceeds normally (fail-open).

Outside allowed time window

The Allowed Search Window field restricts scheduled cycles to one or more time-of-day windows. When the current container-local time falls outside every configured window, the cycle writes one outside allowed time window info row with the current time and the configured windows, then sleeps. Manual Run Now clicks bypass this gate.

Tag filter

Tag Filter · Include and Tag Filter · Exclude in instance settings scope the missing, cutoff, and upgrade passes to (or away from) items carrying specific *arr tags. Both fields take comma-separated tag labels and default to empty. With both empty the filter is a no-op and behavior matches earlier versions.

The engine resolves labels to numeric tag IDs once per cycle by GET-ing each instance's /tag endpoint, so renaming a tag in Radarr or Sonarr does not require re-editing the field. Two cycle-level info rows can appear:

• tag filter (unknown label) lists labels the operator typed that did not resolve to any *arr tag on the current cycle. The remaining labels still apply.
• tag filter (fetch failed) indicates the /tag GET failed for that one cycle. The filter is disabled for that cycle and the search pass proceeds normally; the next cycle retries the fetch.

See Instance Settings > Tag filter for the field reference and the per-app tag-source mapping.

Log deduplication

Eight reasons are deduplicated in the log: on cooldown, on cutoff cooldown, on upgrade cooldown, in hot retry window, waiting on post-release grace, already in download queue, and the two tag filter skip reasons. Each (instance, item, pass, reason) combination writes at most one search_log row every 24 hours on scheduled cycles, except that the two tag filter reasons share one entry and so suppress each other. Run now always writes its rows. The engine still evaluates every candidate every cycle; only the log write is suppressed. This keeps the logs readable when many items share the same cooldown, the same hot-retry interval throttle, or the same tag-filter outcome.

The window lives in memory, in a single cache of 1024 entries shared by every instance, so a restart starts it over. Past 1024 combinations the oldest entries drop out and the suppression stops, which works out to roughly 1024 wanted items with only the missing pass enabled, 341 with all three, or 170 per instance across six. Above that, expect a row per skip per cycle.

The other reasons in the table above write a row every cycle they apply.

Why skips are normal

A high skip count with zero errors is pacing working as designed. The engine evaluates candidates, finds most ineligible, and waits.

Worked example: 500 monitored movies, 50 flagged cutoff-unmet, 35 of those on cooldown, 8 inside post-release grace, batch size 1. The cycle searches 1 movie and skips 49. Over days and weeks the engine works through the backlog as cooldowns expire and grace windows close.

Errors (HTTP 401, connection refused) are the real signal that something is wrong. See Troubleshoot Connection when errors appear.