Full Tunnel low speed #1001

Activity

1. therealaleph commented on May 10, 2026

   

   therealaleph

   on May 10, 2026

   Owner

   More actions

   Reviewed via Anthropic Claude.

   @ehsan272727 — Full mode is structurally slower than apps_script mode for two reasons:

   1. Cache bypass: Full mode tunnels raw TCP. Code.gs's spreadsheet cache (which compounds for repeat assets) doesn't apply because every TCP flow is opaque.
   2. Round-trip stack: client → Apps Script → CodeFull.gs → tunnel-node → destination. Every batch goes through 3 layers minimum.

   For per-batch tuning that helps:

   {
     "coalesce_step_ms": 10,
     "coalesce_max_ms": 1000,
     "parallel_relay": 2
   }

   • parallel_relay: 2 fans across 2 deployments per request — kills long-tail latency at the cost of 2× quota per non-cached request. Add 2-3 script_ids from different Google accounts.
   • coalesce_max_ms higher (up to 5000) packs more ops per batch — saves quota and reduces RTT count, costs initial latency.

   For tunnel-node side: PR #958 (in community testing) batches CacheService.getAll which is a real Full-mode perf win. PR #903 (per-session pipelining) lifts the per-session throughput cap. Both will land in coming weeks if community testing shows no regression.

   Architectural answer: Full mode IS slower than apps_script mode. The trade-off is access to non-Google sites that need the tunnel. If apps_script mode covers your needs (Google, gstatic, googleusercontent — most of the modern web), use that.

   می‌بندم — perf is being actively worked on; track #903 / #958 / #977 for the PR cycle.
2. therealaleph

   closed this as completedon May 10, 2026