VPN: A/B test whether the NordVPN tunnel is degrading streams #118

Closed
opened 2026-11-08 12:10:03 -05:00 by cmoriarty · 2 comments
Owner

Why

Owner, 2026-11-08: no game has played cleanly for more than 5–10 minutes today. The prime suspect is the NordVPN dedicated-IP stack: docker-compose.yml has been the gluetun/WireGuard stack since 6bcfd2b / 02bed23 on develop, live since about 07:00Z today (sb-gluetun, sb-flaresolverr and the app all recreated 16:02Z with #116's deploy).

This test is meant to prove or rule out the VPN before we act on it, because today has other suspects too:

  • For the VPN: the Privado trial on 2026-10-25 was reverted after about 30 minutes. hls.lshls.st sent 800 KB segments slower than their 4 s through both exits (under 2 Mbit/s), while the same tunnel did 27–45 Mbit/s to OVH.
  • Against it, or as a second cause:
    • Today also brought 1.2.0 (08:48Z) and #116's slow-source changes (16:02Z).
    • The Liverpool freezes this afternoon came from hls.lshls.st publishing at 0.86–0.93×, measured from the Mac outside the VPN too.

What to measure

A channel stutters for one of two reasons, and only the second can be the VPN's fault:

  1. The source publishes slower than real time. This is the seconds of video added to its playlist per second of wall time. No tunnel changes it, so it is the control.
  2. Segments arrive slower than they play, or not at all. This is a segment's fetch time ÷ its duration, plus 403s, challenges, decoys and playlist refresh time.

Also count challenges and solve times for the hubs and the solver, since the exit address is new to Cloudflare.

Step 0: look back at today (cheap, confounded)

Using /api/events, compare swaps, recoveries and filler seconds per game-hour before and after ~07:00Z. This is only a hint, because the 08:48Z and 16:02Z deploys fall in the same window.

Test 1: one stream over both paths at the same time (no downtime)

  1. On the VM, run the same probe twice in parallel against one live game's upstream playlist:
    • Path A inside the tunnel (network_mode: container:sb-gluetun).
    • Path B on the VM's own network.
    • The upstream URL and headers are in /app/data/readiness.json; read it with Portainer's archive API.
  2. Poll about once a second for 15+ minutes and record, for each path:
    • the playlist's fetch time
    • the publish rate
    • each segment's bytes, fetch time, Mbit/s and fetch ÷ duration
    • the HTTP status and first bytes (to catch decoys)
  3. Repeat on at least 3 games across the CDNs in use (hls.lshls.st, edgestream*.pro / the hls_co JSON on tiktokcdn, and any others live).
  4. Add a raw throughput baseline: download the same large file over each path. Cloudflare's speed endpoint refused Privado's exit, so use OVH as on 10-25.
  5. Caveat: a stream URL can be bound to the address that cleared the site's challenge (see the compose notes). If path B gets 403s on a link resolved inside the tunnel, resolve the link separately on each path. Record the 403s too, because they answer whether the stream fetches alone could leave outside the tunnel.

Test 2: the whole stack with and without the VPN

  1. Alternate the stacks A-B-A-B over comparable game windows (same sites and CDNs where possible, similar time of day), at least two windows each.
    • Stack 131 deploys docker-compose.yml from develop, so each switch is a commit that swaps in docker-compose.novpn.yml (as fd1d5a6 did), followed by a redeploy.
  2. Main metric (the owner's): the longest clean stretch per game, meaning no filler, splice, swap or recovery.
  3. Also record:
    • swaps, recoveries and filler seconds per game-hour
    • Choppy and Grade.slow grades
    • seconds_since_new_segment gaps over 8 s
    • the Sites table's challenges and solve times
  4. Sources: /api/events, the container log's clock: seq N splices / recovered on lines, and /api/admin.
  5. Swap pitfalls (2026-10-25):
    • Stop and remove sb-gluetun before bringing up the plain stack, because it holds 7000 and 8191.
    • Pass the stack's full Env list back on every redeploy.
    • Switch only when nobody is watching.

Done when

  • Test 1 table, one row per game per path: publish rate, median/p95 fetch ÷ duration, Mbit/s, 403s and decoys.
  • Test 2 table, one row per window: clean-stretch length, swaps, recoveries and filler per game-hour, and challenges.
  • A verdict posted here: the VPN causes it, doesn't, or partly (name which CDN). The stack is left on the winner, with the reason added to the compose notes.
  • If the VPN is cleared, a new issue for the real cause, with these numbers.

If the tunnel is the cause, check next

  • The physical distance from the dedicated IP's exit to the CDN edges (RTT over each path).
  • WireGuard MTU (WIREGUARD_MTU in gluetun) and fragmentation.
  • Whether gluetun is running kernel WireGuard or userspace, and the VM's CPU while it streams.
  • Split routing: keep the hubs and the solver in the tunnel and send only segment fetches direct. This works only if Test 1's path B accepts tunnel-resolved links.
## Why Owner, 2026-11-08: no game has played cleanly for more than 5–10 minutes today. The prime suspect is the NordVPN dedicated-IP stack: `docker-compose.yml` has been the gluetun/WireGuard stack since 6bcfd2b / 02bed23 on develop, live since about 07:00Z today (`sb-gluetun`, `sb-flaresolverr` and the app all recreated 16:02Z with #116's deploy). This test is meant to prove or rule out the VPN before we act on it, because today has other suspects too: - **For the VPN:** the Privado trial on 2026-10-25 was reverted after about 30 minutes. `hls.lshls.st` sent 800 KB segments slower than their 4 s through both exits (under 2 Mbit/s), while the same tunnel did 27–45 Mbit/s to OVH. - **Against it, or as a second cause:** - Today also brought 1.2.0 (08:48Z) and #116's slow-source changes (16:02Z). - The Liverpool freezes this afternoon came from `hls.lshls.st` *publishing* at 0.86–0.93×, measured from the Mac outside the VPN too. ## What to measure A channel stutters for one of two reasons, and only the second can be the VPN's fault: 1. **The source publishes slower than real time.** This is the seconds of video added to its playlist per second of wall time. No tunnel changes it, so it is the control. 2. **Segments arrive slower than they play, or not at all.** This is a segment's fetch time ÷ its duration, plus 403s, challenges, decoys and playlist refresh time. Also count challenges and solve times for the hubs and the solver, since the exit address is new to Cloudflare. ## Step 0: look back at today (cheap, confounded) Using `/api/events`, compare swaps, recoveries and filler seconds per game-hour before and after ~07:00Z. This is only a hint, because the 08:48Z and 16:02Z deploys fall in the same window. ## Test 1: one stream over both paths at the same time (no downtime) 1. On the VM, run the same probe twice in parallel against one live game's upstream playlist: - **Path A** inside the tunnel (`network_mode: container:sb-gluetun`). - **Path B** on the VM's own network. - The upstream URL and headers are in `/app/data/readiness.json`; read it with Portainer's archive API. 2. Poll about once a second for 15+ minutes and record, for each path: - the playlist's fetch time - the publish rate - each segment's bytes, fetch time, Mbit/s and fetch ÷ duration - the HTTP status and first bytes (to catch decoys) 3. Repeat on at least 3 games across the CDNs in use (`hls.lshls.st`, `edgestream*.pro` / the hls_co JSON on tiktokcdn, and any others live). 4. Add a raw throughput baseline: download the same large file over each path. Cloudflare's speed endpoint refused Privado's exit, so use OVH as on 10-25. 5. Caveat: a stream URL can be bound to the address that cleared the site's challenge (see the compose notes). If path B gets 403s on a link resolved inside the tunnel, resolve the link separately on each path. Record the 403s too, because they answer whether the stream fetches alone could leave outside the tunnel. ## Test 2: the whole stack with and without the VPN 1. Alternate the stacks A-B-A-B over comparable game windows (same sites and CDNs where possible, similar time of day), at least two windows each. - Stack 131 deploys `docker-compose.yml` from develop, so each switch is a commit that swaps in `docker-compose.novpn.yml` (as fd1d5a6 did), followed by a redeploy. 2. **Main metric (the owner's):** the longest clean stretch per game, meaning no filler, splice, swap or recovery. 3. **Also record:** - swaps, recoveries and filler seconds per game-hour - Choppy and `Grade.slow` grades - `seconds_since_new_segment` gaps over 8 s - the Sites table's challenges and solve times 4. Sources: `/api/events`, the container log's `clock: seq N splices` / `recovered on` lines, and `/api/admin`. 5. Swap pitfalls (2026-10-25): - Stop and remove `sb-gluetun` before bringing up the plain stack, because it holds 7000 and 8191. - Pass the stack's full Env list back on every redeploy. - Switch only when nobody is watching. ## Done when - [ ] Test 1 table, one row per game per path: publish rate, median/p95 fetch ÷ duration, Mbit/s, 403s and decoys. - [ ] Test 2 table, one row per window: clean-stretch length, swaps, recoveries and filler per game-hour, and challenges. - [ ] A verdict posted here: the VPN causes it, doesn't, or partly (name which CDN). The stack is left on the winner, with the reason added to the compose notes. - [ ] If the VPN is cleared, a new issue for the real cause, with these numbers. ## If the tunnel is the cause, check next - The physical distance from the dedicated IP's exit to the CDN edges (RTT over each path). - WireGuard MTU (`WIREGUARD_MTU` in gluetun) and fragmentation. - Whether gluetun is running kernel WireGuard or userspace, and the VM's CPU while it streams. - Split routing: keep the hubs and the solver in the tunnel and send only segment fetches direct. This works only if Test 1's path B accepts tunnel-resolved links.
Author
Owner

Verdict (2026-11-08, 17:50–18:55Z)

The tunnel doesn't slow streams. The VPN stack hurts in two other ways:

  1. gluetun's DNS filter (BLOCK_MALICIOUS, on by default) answers REFUSED for the hosts of most players and of two of the fastest stream CDNs.
    • Hosts refused: streame.center, embedcenter.st, edgestream3.pro, hls.meth-streams-hd.cfd, s1.saltext.ch, play.aurasportsplayer.cfd, frame.embedme.cfd, and ad/analytics hosts.
    • The same names resolve at 1.1.1.1 asked through the same tunnel, so this is gluetun's list, not NordVPN.
    • It is also where the new manifest: upstream unreachable: ConnectError link failures come from (iohit.st, shd247.world): the stream host never resolves.
  2. edgestream3.pro's player refuses the NordVPN address even with DNS fixed, and its links (and s1.saltext.ch's) are tied to the address that opened the player.

The result:

  • Through the VPN the app finds streams on very few sites, and what it has left is mostly hls.lshls.st, which is slow at peak on any path.
  • Direct, the same sites lead to edgestream3.pro and hls.meth-streams-hd.cfd, which deliver a segment in 4–10% of its play time.

How it was measured

Two short-lived probe containers on the VM ran the app's own image, fetch code and browser code. One shared sb-gluetun's network (tunnel, exit Ashburn VA); the other was on Docker's bridge (direct, the home address). They tested the same targets at the same moment, with no downtime, and were removed afterwards.

The tunnel itself is healthy

Tunnel Direct
WireGuard in-kernel, MTU 1440 by path-MTU discovery, no health restarts —
One connection, 100 MB (Hetzner Ashburn / OVH Canada / Cloudflare) 576 / 546 / 508 Mbit/s 757 / 555 / 750 Mbit/s
4 connections, Hetzner Ashburn 589 Mbit/s 730 Mbit/s
TCP+TLS to hls.lshls.st / instreams.pro / topsurge.st / powerstream.club / shd247.world 308 / 216 / 198 / 483 / 300 ms 301 / 189 / 161 / 499 / 257 ms
DNS lookup, 35 site and stream hosts 0.2–53 ms 21–278 ms

Through the tunnel a TCP connect answers in about 14 ms whatever the destination, as if NordVPN's server accepts connections itself and relays them. The TLS handshake is the real end-to-end figure, and it is close on both paths.

hls.lshls.st, 20 minutes at the 1 PM kickoff (18:02–18:22Z)

Every newly listed segment was fetched on each path at the same moment.

Stream Path Segments Publish rate Fetch ÷ play time p50 / p95 / max Slower than real time Failed fetches
Man Utd–Aston Villa (topsurge.st) direct 548 1.0× 1.28 / 2.75 / 4.8 82% 16
tunnel 557 1.0× 1.32 / 3.09 / 7.3 84% 7
Saints–Browns (powerstream.club) direct 309 1.0× 1.24 / 2.56 / 6.1 75% 7
tunnel 313 1.0× 1.28 / 3.09 / 11.0 77% 3

The same server, the same slowness. The tunnel adds about 3% at the median and a heavier tail. (The Saints link expired, with 403s on both paths at once.)

Player pages, 33 sites, read once each from both paths at the same moment

This used the app's own browser code, before any solver handoff.

Stream found Page loaded, no stream Challenged
Tunnel 3 25 5
Direct 15 14 4
  • Found direct, nothing through the tunnel (13 sites): givemestreams.cfd, sportsframe.site, resports.cfd, s2watch.cyou, bally-sports.lat, iptvcast.lol, buff-streams.buzz, wildgames.bz, fast-streams.biz, shd247.world, nfl-castnow.lat, watchsports.online, and ihdstreams.top (which was challenged through the tunnel).
  • The other way round: only topsurge.st.
  • Challenged on both paths: streameast.ga, tvsportslive.fr, dudestream1.com and suksukstreams.cfd. They challenge a headless browser whatever its address.

With the refused names looked up at 1.1.1.1 inside the tunnel:

  • the 6 meth-streams sites load their playlist through the VPN;
  • the 7 edgestream3.pro sites still show no stream.

The owner's split: find at home, play through the VPN

Sixteen streams were found from home, then fetched for 2–3 minutes. The tunnel side had its DNS fixed the same way.

Stream server Sites Direct: fetch ÷ play time p50 (max) Through the VPN
edgestream3.pro 7 0.07–0.10 (0.5–1.3) 403 every time
hls.meth-streams-hd.cfd 5 (+1 dead link) 0.04–0.05 (0.45–0.65) plays: 0.06 (0.19–0.31)
s1.saltext.ch 1 0.25 (0.37) 403 every time
hls.lshls.st 2 0.79–0.80 (1.1) plays: 0.68–0.73 (2.3)

Step 0: today's history (confounded by 1.2.0, #116, #119 and the game mix)

  • Channels: 1.3 swaps and 3.8 recoveries per channel-hour since about 07:00Z, against 4.1 and 11.0 before. The median longest clean stretch is 12.1 minutes, against 6.7. Playback itself isn't worse.
  • Checks: a check found a stream 94% of the time before (63 of 67), and 54% since (30 of 56).
  • New since the switch: 14 links failed on "upstream unreachable: ConnectError" (DNS refused), and there were 13 "challenged the app's browser" events, against none before.
  • Live at 18:23Z: 8 followed live games, 1 with a stream.

Options

Option Change Today's 16 sites with a stream Fast CDNs What sees the home address
Now (VPN, filter on) — ~2–3, mostly hls.lshls.st none nothing
A. VPN, DNS filter off BLOCK_MALICIOUS=off on gluetun, redeploy about 8–9: hls.lshls.st 2, meth-streams 5, saltext 1 (if found through the VPN, whose playlist loaded) meth-streams nothing
B. Owner's split: find at home, play through the VPN, filter off compose + one app setting (gluetun's HTTP proxy for stream fetches) 7: hls.lshls.st 2, meth-streams 5; edgestream3 and saltext get 403 meth-streams player pages, hubs, the solver
B+. Split, with direct play when the VPN gets a 403 as B + fallback logic 16 all the above, plus edgestream3/saltext video
C. No VPN swap in the plain compose 16 all everything
  • On B: finding at home helps less than expected. In the same-moment test, challenges were nearly equal (4 against 5), and the day's history (0 against 13) suggests the VPN address draws more only over time.
  • B's real catch: the fastest CDN today, edgestream3.pro, won't play a home-found link through the VPN. So B alone covers fewer sites than A, and B+ plays those streams from home anyway.

Recommendation:

  1. A, now. It is one line, keeps everything in the VPN, and roughly triples the usable links, adding the fastest CDN that accepts the VPN.
  2. Then the owner's call between privacy and coverage: C (or B+) gains edgestream3.pro's 7 sites by exposing the home address to them.

Test 2 (the whole-stack A/B) wasn't run: the same-moment tests above show the cause directly. After A, one game window will confirm it, measured by the longest clean stretch, swaps, recoveries and filler per game-hour, and the check success rate.

## Verdict (2026-11-08, 17:50–18:55Z) **The tunnel doesn't slow streams. The VPN stack hurts in two other ways:** 1. **gluetun's DNS filter** (`BLOCK_MALICIOUS`, on by default) answers REFUSED for the hosts of most players and of two of the fastest stream CDNs. - Hosts refused: `streame.center`, `embedcenter.st`, `edgestream3.pro`, `hls.meth-streams-hd.cfd`, `s1.saltext.ch`, `play.aurasportsplayer.cfd`, `frame.embedme.cfd`, and ad/analytics hosts. - The same names resolve at 1.1.1.1 asked **through the same tunnel**, so this is gluetun's list, not NordVPN. - It is also where the new `manifest: upstream unreachable: ConnectError` link failures come from (iohit.st, shd247.world): the stream host never resolves. 2. **edgestream3.pro's player refuses the NordVPN address** even with DNS fixed, and its links (and `s1.saltext.ch`'s) are tied to the address that opened the player. **The result:** - Through the VPN the app finds streams on very few sites, and what it has left is mostly `hls.lshls.st`, which is slow at peak on any path. - Direct, the same sites lead to `edgestream3.pro` and `hls.meth-streams-hd.cfd`, which deliver a segment in 4–10% of its play time. ## How it was measured Two short-lived probe containers on the VM ran the app's own image, fetch code and browser code. One shared `sb-gluetun`'s network (tunnel, exit Ashburn VA); the other was on Docker's bridge (direct, the home address). They tested the same targets at the same moment, with no downtime, and were removed afterwards. ### The tunnel itself is healthy | | Tunnel | Direct | |---|---|---| | WireGuard | in-kernel, MTU 1440 by path-MTU discovery, no health restarts | — | | One connection, 100 MB (Hetzner Ashburn / OVH Canada / Cloudflare) | 576 / 546 / 508 Mbit/s | 757 / 555 / 750 Mbit/s | | 4 connections, Hetzner Ashburn | 589 Mbit/s | 730 Mbit/s | | TCP+TLS to hls.lshls.st / instreams.pro / topsurge.st / powerstream.club / shd247.world | 308 / 216 / 198 / 483 / 300 ms | 301 / 189 / 161 / 499 / 257 ms | | DNS lookup, 35 site and stream hosts | 0.2–53 ms | 21–278 ms | Through the tunnel a TCP connect answers in about 14 ms whatever the destination, as if NordVPN's server accepts connections itself and relays them. The TLS handshake is the real end-to-end figure, and it is close on both paths. ### hls.lshls.st, 20 minutes at the 1 PM kickoff (18:02–18:22Z) Every newly listed segment was fetched on each path at the same moment. | Stream | Path | Segments | Publish rate | Fetch ÷ play time p50 / p95 / max | Slower than real time | Failed fetches | |---|---|---|---|---|---|---| | Man Utd–Aston Villa (topsurge.st) | direct | 548 | 1.0× | 1.28 / 2.75 / 4.8 | 82% | 16 | | | tunnel | 557 | 1.0× | 1.32 / 3.09 / 7.3 | 84% | 7 | | Saints–Browns (powerstream.club) | direct | 309 | 1.0× | 1.24 / 2.56 / 6.1 | 75% | 7 | | | tunnel | 313 | 1.0× | 1.28 / 3.09 / 11.0 | 77% | 3 | The same server, the same slowness. The tunnel adds about 3% at the median and a heavier tail. (The Saints link expired, with 403s on both paths at once.) ### Player pages, 33 sites, read once each from both paths at the same moment This used the app's own browser code, before any solver handoff. | | Stream found | Page loaded, no stream | Challenged | |---|---|---|---| | Tunnel | 3 | 25 | 5 | | Direct | 15 | 14 | 4 | - **Found direct, nothing through the tunnel (13 sites):** givemestreams.cfd, sportsframe.site, resports.cfd, s2watch.cyou, bally-sports.lat, iptvcast.lol, buff-streams.buzz, wildgames.bz, fast-streams.biz, shd247.world, nfl-castnow.lat, watchsports.online, and ihdstreams.top (which was challenged through the tunnel). - **The other way round:** only topsurge.st. - **Challenged on both paths:** streameast.ga, tvsportslive.fr, dudestream1.com and suksukstreams.cfd. They challenge a headless browser whatever its address. With the refused names looked up at 1.1.1.1 inside the tunnel: - the 6 meth-streams sites load their playlist through the VPN; - the 7 edgestream3.pro sites still show no stream. ### The owner's split: find at home, play through the VPN Sixteen streams were found from home, then fetched for 2–3 minutes. The tunnel side had its DNS fixed the same way. | Stream server | Sites | Direct: fetch ÷ play time p50 (max) | Through the VPN | |---|---|---|---| | edgestream3.pro | 7 | 0.07–0.10 (0.5–1.3) | **403 every time** | | hls.meth-streams-hd.cfd | 5 (+1 dead link) | 0.04–0.05 (0.45–0.65) | plays: 0.06 (0.19–0.31) | | s1.saltext.ch | 1 | 0.25 (0.37) | **403 every time** | | hls.lshls.st | 2 | 0.79–0.80 (1.1) | plays: 0.68–0.73 (2.3) | ### Step 0: today's history (confounded by 1.2.0, #116, #119 and the game mix) - **Channels:** 1.3 swaps and 3.8 recoveries per channel-hour since about 07:00Z, against 4.1 and 11.0 before. The median longest clean stretch is 12.1 minutes, against 6.7. Playback itself isn't worse. - **Checks:** a check found a stream 94% of the time before (63 of 67), and 54% since (30 of 56). - **New since the switch:** 14 links failed on "upstream unreachable: ConnectError" (DNS refused), and there were 13 "challenged the app's browser" events, against none before. - **Live at 18:23Z:** 8 followed live games, 1 with a stream. ## Options | Option | Change | Today's 16 sites with a stream | Fast CDNs | What sees the home address | |---|---|---|---|---| | **Now** (VPN, filter on) | — | ~2–3, mostly hls.lshls.st | none | nothing | | **A.** VPN, DNS filter off | `BLOCK_MALICIOUS=off` on gluetun, redeploy | about 8–9: hls.lshls.st 2, meth-streams 5, saltext 1 (if found through the VPN, whose playlist loaded) | meth-streams | nothing | | **B.** Owner's split: find at home, play through the VPN, filter off | compose + one app setting (gluetun's HTTP proxy for stream fetches) | 7: hls.lshls.st 2, meth-streams 5; edgestream3 and saltext get 403 | meth-streams | player pages, hubs, the solver | | **B+.** Split, with direct play when the VPN gets a 403 | as B + fallback logic | 16 | all | the above, plus edgestream3/saltext video | | **C.** No VPN | swap in the plain compose | 16 | all | everything | - **On B:** finding at home helps less than expected. In the same-moment test, challenges were nearly equal (4 against 5), and the day's history (0 against 13) suggests the VPN address draws more only over time. - **B's real catch:** the fastest CDN today, edgestream3.pro, won't play a home-found link through the VPN. So B alone covers fewer sites than A, and B+ plays those streams from home anyway. **Recommendation:** 1. **A, now.** It is one line, keeps everything in the VPN, and roughly triples the usable links, adding the fastest CDN that accepts the VPN. 2. **Then the owner's call between privacy and coverage:** C (or B+) gains edgestream3.pro's 7 sites by exposing the home address to them. Test 2 (the whole-stack A/B) wasn't run: the same-moment tests above show the cause directly. After A, one game window will confirm it, measured by the longest clean stretch, swaps, recoveries and filler per game-hour, and the check success rate.
Author
Owner

After turning gluetun's DNS filter off (PR #120, merge 7700913, deployed 19:02:44Z)

The owner chose option A. The probes were then repeated, and the app was watched on its own.

The same page test, 33 player sites, both paths at the same moment

Run Stream found Page loaded, no stream Challenged
Tunnel, filter on (18:22Z) 3 25 5
Direct (18:22Z) 15 14 4
Tunnel, filter off (19:03Z) 16 13 4
Direct (19:03Z) 16 13 4

The tunnel now matches direct site for site.

Correction to my earlier report: the 7 edgestream3.pro sites do give streams through the VPN. "edgestream3's player refuses the NordVPN address" was wrong: it came from my test's DNS workaround, not from the site. So option A covers everything that works from home, and options B, B+ and C are moot for coverage.

Playback through the tunnel of streams found through the tunnel (3 minutes, 19:11Z)

Stream server Sites Fetch ÷ play time p50 (max) Segments slower than real time
edgestream3.pro 6 0.07–0.09 (0.57–1.22) 1 of 276
hls.meth-streams-hd.cfd 4 0.07 (0.45–0.81) 0
s1.saltext.ch 1 0.28 (0.45) 0
hls.lshls.st 2 0.56 (4.1–4.2) 28 of 194

The app's own numbers

Window Checks that found a stream Links verified DNS/connect link failures Live followed games with a stream
17:29–19:02Z, filter on 20 of 38 12 of 146 (8%) 7 1 of 7 (19:00Z)
19:18–19:38Z, filter off, no probes 11 of 11 5 of 21 (9 more were copies of streams already kept) 1 4 of 8 (19:38Z)
  • At 19:38Z: the four channels were warm on edgestream3.pro and s1.saltext.ch, all Healthy, each holding its full 16–24 s head start (#117). Three of them had no swap, recovery or filler in the window.
  • Failures in those 20 minutes: a connection timeout at 19:18:51, 30 s after the probes stopped; a source's 503s; and one degrading swap.

What the tunnel still costs: connection setup under bursts

480 fresh TCP+TLS connections, 4 at a time to each of 4 stream servers Direct Tunnel
Timed out (5 s) 0 10 (2.1%)
p95 setup time 250–470 ms 1.3–3.4 s
  • One connection at a time through the tunnel is clean: 239 of 240, p95 243–354 ms. Bursts to nearby Cloudflare are clean too. The loss shows only under bursts to the far servers.
  • When it bites: from 19:13 to 19:17Z, while my probes added about 13 streams and 480 burst connections to the same tunnel, several channels failed at once on "no new segment … ConnectTimeout". With the app alone it happened once in 20 minutes.
  • Why the app is exposed: it opens a fresh connection for every playlist poll and segment (upstream._client() per request). That load includes the watch's polls of up to 24 links.
  • The fix: reusing connections (keep-alive, one pooled client) would take most of those connections off the tunnel. That is a follow-up worth filing.

Verdict

Partly the VPN stack, not the tunnel.

  • The cause of today's poor coverage was gluetun's DNS filter. It refused the players and CDNs that edgestream3.pro, hls.meth-streams-hd.cfd and s1.saltext.ch streams need, and that left the app on hls.lshls.st, which is slow at peak on any path.
  • With the filter off (#120), the VPN finds every stream the home connection does, and plays them as fast.
  • What remains is a small connection-setup cost under bursts.

The stack stays on the VPN with the filter off; the reason is in the compose notes (#120). Closing, as the fix is merged.

## After turning gluetun's DNS filter off (PR #120, merge 7700913, deployed 19:02:44Z) The owner chose option A. The probes were then repeated, and the app was watched on its own. ### The same page test, 33 player sites, both paths at the same moment | Run | Stream found | Page loaded, no stream | Challenged | |---|---|---|---| | Tunnel, filter on (18:22Z) | 3 | 25 | 5 | | Direct (18:22Z) | 15 | 14 | 4 | | **Tunnel, filter off (19:03Z)** | **16** | 13 | 4 | | Direct (19:03Z) | 16 | 13 | 4 | The tunnel now matches direct site for site. **Correction to my earlier report:** the 7 edgestream3.pro sites do give streams through the VPN. "edgestream3's player refuses the NordVPN address" was wrong: it came from my test's DNS workaround, not from the site. So option A covers everything that works from home, and options B, B+ and C are moot for coverage. ### Playback through the tunnel of streams found through the tunnel (3 minutes, 19:11Z) | Stream server | Sites | Fetch ÷ play time p50 (max) | Segments slower than real time | |---|---|---|---| | edgestream3.pro | 6 | 0.07–0.09 (0.57–1.22) | 1 of 276 | | hls.meth-streams-hd.cfd | 4 | 0.07 (0.45–0.81) | 0 | | s1.saltext.ch | 1 | 0.28 (0.45) | 0 | | hls.lshls.st | 2 | 0.56 (4.1–4.2) | 28 of 194 | ### The app's own numbers | Window | Checks that found a stream | Links verified | DNS/connect link failures | Live followed games with a stream | |---|---|---|---|---| | 17:29–19:02Z, filter on | 20 of 38 | 12 of 146 (8%) | 7 | 1 of 7 (19:00Z) | | 19:18–19:38Z, filter off, no probes | 11 of 11 | 5 of 21 (9 more were copies of streams already kept) | 1 | **4 of 8 (19:38Z)** | - **At 19:38Z:** the four channels were warm on edgestream3.pro and s1.saltext.ch, all Healthy, each holding its full 16–24 s head start (#117). Three of them had no swap, recovery or filler in the window. - **Failures in those 20 minutes:** a connection timeout at 19:18:51, 30 s after the probes stopped; a source's 503s; and one degrading swap. ### What the tunnel still costs: connection setup under bursts | 480 fresh TCP+TLS connections, 4 at a time to each of 4 stream servers | Direct | Tunnel | |---|---|---| | Timed out (5 s) | 0 | 10 (2.1%) | | p95 setup time | 250–470 ms | 1.3–3.4 s | - **One connection at a time** through the tunnel is clean: 239 of 240, p95 243–354 ms. Bursts to nearby Cloudflare are clean too. The loss shows only under bursts to the far servers. - **When it bites:** from 19:13 to 19:17Z, while my probes added about 13 streams and 480 burst connections to the same tunnel, several channels failed at once on "no new segment … ConnectTimeout". With the app alone it happened once in 20 minutes. - **Why the app is exposed:** it opens a fresh connection for every playlist poll and segment (`upstream._client()` per request). That load includes the watch's polls of up to 24 links. - **The fix:** reusing connections (keep-alive, one pooled client) would take most of those connections off the tunnel. That is a follow-up worth filing. ## Verdict **Partly the VPN stack, not the tunnel.** - The cause of today's poor coverage was gluetun's DNS filter. It refused the players and CDNs that `edgestream3.pro`, `hls.meth-streams-hd.cfd` and `s1.saltext.ch` streams need, and that left the app on `hls.lshls.st`, which is slow at peak on any path. - With the filter off (#120), the VPN finds every stream the home connection does, and plays them as fast. - What remains is a small connection-setup cost under bursts. The stack stays on the VPN with the filter off; the reason is in the compose notes (#120). Closing, as the fix is merged.
Sign in to join this conversation.
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/sports-breaker#118
No description provided.