Connection warming1.x

Connection warming since 1.2.0 keeps a small number of pre-connected, authenticated NNTP sockets standing by on each pooled provider. Playback and queue work borrow one instantly instead of paying the TCP + TLS + login handshake first — which is most of the wait between pressing play and the first bytes moving after an idle period.

How it works

Each pooled provider gets a warm floor: a target count of idle, ready-to-use connections. By default the floor is 2 (1 when the Provider Connection Limit is 1) since 1.5.0; releases before 1.5.0 derived roughly one sixth of the limit, clamped to 1–8. An explicit usenet.warm-connections.floor value is unchanged.

1. In the background

   1. SweeperRuns every half idle timeout, 30 seconds by default
   2. DATE pingKeeps each warm socket alive before the provider drops it
   3. RefillReplaces stale sockets and closes idle ones above the floor

2. When playback or the queue needs a connection

   1. Playback, queueAsk the pool for a connection
   2. Warm floorHands over a socket that is already logged in
   3. Usenet providerGets the first request with no TCP, TLS or login wait

• A background sweeper runs every half of the idle connection timeout (default: every 30 seconds). It refills the floor when sockets were lost and reaps idle connections only above the floor — warm connections are never closed just for being idle.
• Before a provider’s own idle timeout can drop a warm socket, the sweeper pings it with a lightweight NNTP DATE command. A failed ping means the socket went stale; it is disposed and the floor refills.
• Warm connections never hold download permits, so the full configured connection width stays available to real work.
• Disabled providers are never warmed since 1.5.0: infinidysk opens no connections to a disabled account and sends no keep-alives until the provider is enabled again.
• Backup-only providers are demand-only since 1.6.1: they keep no warm floor and receive no keep-alives, so a last-resort article-fill account is not dialed in the background. Failover still opens a connection when a higher-priority provider misses an article; that first fallback pays the handshake, and the connection is reused until it goes idle. Backup + Stats providers keep their floor.
• If a provider is unreachable at startup, warming does not spin or block startup — retries resume once the provider circuit permits new opens. When infinidysk learns that a provider’s real connection limit is lower than configured, the floor shrinks with it.
• Pending warm-up work has a 15-second acquisition budget and stops when its provider circuit trips. Factories already opening keep their own socket-open deadline. Warm-up never claims the half-open recovery probe.
• Warm-up opens are judged differently from foreground opens since 1.5.0. While a provider still has at least one established connection, a warm-up open that fails or times out never trips the circuit: the pool backs off its own handshakes (500 ms doubling to 60 seconds), the floor refills on a later sweep, and a warning naming the reason is logged at most once every five minutes per provider. Only a warm-up failure on a provider with no established connections reaches the circuit, so an unreachable provider is still benched before the first request needs it. Floor refills use a connection-open budget of at least 10 seconds (the configured fresh connection open timeout when that is longer); read-start warm-up keeps the configured budget because it shares handshake slots with playback. See provider acquisition waits.

What you see in the header

The Connections indicator in the top navigation shows pooled-provider totals, updated live (about five times per second):

• 5/50 — live connections across pooled providers versus their combined maximum.
• 2 active · 3 warm — connections currently transferring versus idle warm ones. The warm part only appears while at least one warm connection is standing by.
• Hovering explains: “Warm connections are pre-connected to your Usenet providers so playback can start faster.”

Other states: a spinner with Connecting until the first update arrives, Reconnecting while the live UI connection recovers from a brief drop, and — / No providers when no Usenet provider is configured. The indicator is hidden on narrow screens. Backup-only providers are excluded from the totals.

Settings → Usenet shows the same live counts on each provider card.

Tuning

Setting	What it does
Warm connections usenet.warm-connections.enabled	Keep a small pool of pre-connected sockets per provider. Default on.
Warm floor usenet.warm-connections.floor	Idle sockets kept ready per provider, capped at the Provider Connection Limit. Before 1.5.0 the default derived one sixth of the limit, clamped to 1–8. Default 2.
Read-start warm-up since 1.3.0 usenet.read-start-warmup.enabled	For a long buffered read, open missing pooled-provider connections in parallel as the first segment starts. Default on.

There is no settings-UI toggle; set the keys as headless environment variables (NZBDAV_CONFIG__USENET__WARM_CONNECTIONS__ENABLED / NZBDAV_CONFIG__USENET__WARM_CONNECTIONS__FLOOR). Changes take effect on the next provider save or restart — connection pools are not rebuilt when these keys change alone.

Read-start warm-up is enabled by default. Each newly opened WebDAV stream snapshots the value. Reads below about 8.4 MB, reads needing fewer than two connection batches, backup-only providers, providers with an open circuit, and providers that are backing off after a failed connection open since 1.5.0 are skipped. The target is bounded by the stream’s article window, download limit, provider connection limits, and any per-provider transfer cap. Opened sockets return to the normal idle pool; the warm floor still decides how many remain connected later. Set NZBDAV_CONFIG__USENET__READ_START_WARMUP__ENABLED=false and restart to restore demand-only expansion.

Cost and trade-offs

• Warm sockets are real connections and count against the provider plan’s connection limit, even while idle.
• Read-start warm-up can create up to three TCP/TLS/AUTHINFO sessions at once and pauses by itself while a provider is backing off after a failed open. Turn it off if a provider rejects legal connection bursts.
• Keepalive traffic is negligible: one tiny DATE exchange per warm socket per sweep.
• On very small plans or memory-constrained hosts, lower the floor or disable warming entirely.

Usenet · NNTP pipelining · Streaming and seeking