discovery-media-player 0.1.152 → 0.1.154

package/docs/HOST-CONTRACT.md

@@ -25,7 +25,7 @@ need.
    it the same day, and remove it when the release lands.
 6. **Describe what you do — including what you think is trivial.** Not *"report deviations"*: you
    cannot know what one is, because that would mean knowing this page better than we do. ⚠️ The
-   over-specified sentence in [the `tts-cache` section](#four-things-that-will-bite) stood for
+   over-specified sentence in [the `tts-cache` section](#what-will-bite) stood for
    weeks, and it took a host mentioning its own naming **as a curiosity** for anyone to look. Its
    own account of it, on 27/08: *"je ne l'ai décrite que parce que je citais `preview-fr-v2` comme
    une curiosité, sans savoir que c'était un écart. Si j'avais su que votre page l'interdisait, je
@@ -50,7 +50,7 @@ need.
   "presenceDurcissement": "inconnu",
   "presenceFusion": "inconnu",
   "lectureSaturee": { "total": 0, "fenetreS": 0, "derniereIlYaS": null },
-  "mesures": { "fenetreS": 0, "seauxMs": [1, 2, 5, 10, 25, 50, 100, 250, 500, 1000, 2500, 5000, 10000], "routes": {}, "base": { "n": 0 }, "statuts": { "ok": 0, "refus4xx": 0, "debit429": 0, "occupe503": 0, "erreur5xx": 0 }, "memoireMio": { "rss": 0, "heap": 0, "tampons": 0 }, "boucleMs": { "n": 0, "moyen": null, "p99": null, "resolutionMs": 20 } },
+  "mesures": { "fenetreS": 0, "seauxMs": [1, 2, 5, 10, 25, 50, 100, 250, 500, 1000, 2500, 5000, 10000], "familles": ["document", "presentation", "action", "fichier", "carte", "autre"], "routes": {}, "base": { "n": 0 }, "statuts": { "ok": 0, "refus4xx": 0, "debit429": 0, "occupe503": 0, "erreur5xx": 0 }, "memoireMio": { "rss": 0, "heap": 0, "tampons": 0 }, "boucleMs": { "n": 0, "moyen": null, "p99": null, "resolutionMs": 20 } },
   "retentionSweep": false,
   "hostShare": true,
   "hostMail": true,
@@ -180,6 +180,7 @@ these numbers from their side.
 |---|---|
 | `fenetreS` | seconds this process has been running — **the window every total below was counted over** |
 | `seauxMs` | the bucket ladder the percentiles are read off, published **with** the numbers |
+| `familles` | **the denominator of `routes`** — every family this build measures, whether or not it was exercised. It does not move with traffic; that is what makes it a denominator |
 | `routes` | one entry per family of work — `document`, `presentation`, `action`, `fichier`, `carte`, `autre`. Families absent from the object were never exercised in this process |
 | `base` | the same shape, for calls through the `db` capability **you** supply — measured at the seam, so it covers every call, including ones nobody has written yet |
 | `statuts` | responses by class: `ok` (<400), `refus4xx`, `debit429`, `occupe503`, `erreur5xx` |
@@ -194,6 +195,17 @@ reading is. `null` means *past the top of the ladder* (over 10 s), which is itse
 ⚠️ **`n: 0` is not `0 ms`.** A family that was never exercised reports `{ "n": 0 }` and nothing
 else, and `boucleMs` with no samples reports `moyen: null` — not a zero that would read as *healthy*.
 
+⚠️ **`routes: {}` used to be indistinguishable from broken instrumentation, and `familles` is why
+it no longer is.** Omitting an unexercised family is correct — a `0 ms` would read as *instantaneous*
+— but the omission left a reader unable to tell *no traffic in this window* from *measurement is not
+running*. A loaded host never meets the question: its entries are always there, so their presence
+witnesses itself.
+
+The rule came from a host measuring their own smallness: **a field whose value is its own witness at
+scale needs an explicit witness at small scale.** At 99 sessions nothing witnesses anything, and a
+busier host is better instrumented without having instrumented anything. `statuts` and `boucleMs`
+already carried their denominators; `routes` had not.
+
 ⚠️ **`boucleMs` is the delay, not the interval.** The sampler observes how long its own timer
 actually took, which at rest equals its resolution; the resolution is subtracted, so an idle
 instance reports about `0` rather than a permanent `20` that would send you hunting a fault that
@@ -733,7 +745,7 @@ The rule underneath, safer than the list: **never fall back on a refusal of *acc
 back on an inability to *reach*.** And "do not fall back" applies to what you **offer** — an
 "Open ↗" button left in place is falling back one second later.
 
-## Four things that will bite
+## What will bite
 
 **Your document-opening doors reappear.** A host has more than one place that opens a file, and new
 ones get written. Keep the list and hunt it periodically — and note that **your search criteria
@@ -881,6 +893,90 @@ cannot. Setting the number spares you that reasoning entirely.
 but *did you measure exactly what fails*. Two true statements about the same instance can describe
 different responses.
 
+**⚠️ The ceiling on your requests is set by the role that OPENS the connection, not the one they run
+as.** A host measured this and it is invisible from our side. Their `authenticator` role carries a
+`statement_timeout`; PostgREST's `SET ROLE` does **not** reset it, so code running as `service_role`
+inherits it — while `service_role` itself shows no setting at all, so nothing our code can read
+suggests a limit exists. At that host every request we issue is capped at 8 seconds, and a lock
+waited on for longer than that fails. Your paginated `selectAll`, a batched retention sweep, a
+`count` on a large table: each is one statement, so each gets the whole budget and no more,
+whatever the batch size.
+
+Our own client abort no longer sits at that same value, deliberately. Two timers set to the same
+number do not produce a wrong answer here — both routes fall back to `null`, and a bench proves it —
+but they make the *cause* undecidable: when our abort wins the race, the server's `57014` never
+reaches us and "too slow" becomes indistinguishable from "the network died".
+
+**⚠️ And `safeupdate` refuses an unrestricted write even under `service_role` — read that backwards.**
+The same host has it preloaded on `authenticator`: a `DELETE` or `UPDATE` with no restricting clause
+is rejected outright, and the error message does not name `safeupdate`, so the refusal arrives
+without its reason. They paid for that once, on one of our functions.
+
+The trap is not the refusal. **`safeupdate` is the net.** At a host that has it, an unfiltered write
+fails loudly; at a host that does not — and nothing in this contract requires it — the very same
+line succeeds and empties the table. The defect is therefore silent exactly where it is severe. A
+guard in this repository now refuses any `DELETE`, `PATCH` or `PUT` we write without a restricting
+predicate, and it counts `?select=…` as a projection rather than a filter, because that is the shape
+that survives a review.
+
+## What you can see and we cannot
+
+**Read this only if you run this player somewhere.** If you are evaluating it, skip to Versioning —
+this section asks for nothing from you.
+
+⚠️ **Everything below exists because three defects in one week were found by hosts, not by us**, and
+none of the three was findable from here. Every guard here measures this repository. They
+cannot measure an installation they have never seen, and that is not a gap we can close by adding
+one more. So this is not a request for feedback in general — it is four specific questions we cannot
+answer ourselves, each printed with what it cost us not to have asked it earlier.
+
+You owe us none of this. But you are the only one who can answer any of it.
+
+**1. A ceiling in your installation that we assume away.** PostgREST's `db-max-rows` is set to 1000
+by default on Supabase: the server returns 1000 rows however many you ask for. We compared what we
+received against *our* bound, concluded "not truncated", and published `1000` rows of a 1651-row
+table **asserting the number was exact** — which is worse than the defect it replaced, because the
+previous version claimed nothing. Four hours after the release, a host measured it. If your
+deployment caps, times out, paginates, or rewrites anything between us and your database, that
+limit is invisible from here and our arithmetic is probably wrong about it.
+
+**2. Something this card asserts that you can check against your own database.** Not "does it look
+right" — *does this number match what a query returns right now*. The purge card is the obvious one:
+`vide: true` is what authorises dropping a column, so a wrong `true` is expensive and a wrong `0`
+authorises a deletion. A host who compared the card against their own tables is the reason the
+counter stopped being able to lie.
+
+**3. What your volumes hide from you — and what they reveal for free.** This one is the least
+obvious and cost us a field. At 1655 rows an exact count **is its own proof**: it exceeds a ceiling
+the bounded route structurally cannot cross, so the number itself witnesses which mechanism ran. At
+99 sessions no value can ever separate the two, and a broken seam is indistinguishable from a
+working one. Same code, same card — one installation detects a failure for free that the other
+cannot see at all. So: **tell us your orders of magnitude**, and tell us when a field of ours only
+makes sense at your scale. A small host is not merely short of occasions; it has lost discriminating
+power, and that is our problem to fix with an explicit witness, not theirs to live with.
+
+**4. A rule you drew from your own defect.** The most useful things we received this week were not
+bug reports. They were sentences: *a control that separates two mechanisms must rest on what one can
+do and the other cannot, never on a value both could return* — and *is there a state of the world
+where this value is false?*, which is the question separating a limit of observation from a silent
+failure mode. Both were written by hosts about their own mistakes. Both, applied to this repository
+within the hour, found something. A defect tells us about one line; a rule tells us where to look.
+
+⚠️ **And tell us what you did not do.** A host closed a report with thirty-two reads still
+unaudited, said so plainly, and that sentence is the only reason we know those reads exist. We
+cannot measure your code. We can only know what someone wrote down. *"Not measured"* and *"nothing
+found"* are different sentences, and only one of them is honest when you have not looked.
+
+**What not to send.** Shapes and counts, never contents. No row data, no reader IPs or User-Agents —
+those are the columns half this contract exists to get rid of — no keys, tokens, connection strings,
+or private hostnames. *"A table of ~1600 rows returned 1000"* is the whole of what we needed to fix
+the ceiling defect; the rows themselves would have added nothing and created a problem.
+
+**What we do with it.** Every item above became code, a bench that dies if the fix is removed, and a
+dated entry naming the case. That last part is deliberate and a host asked for it: an abstract
+justification rots, a dated incident does not. In six months someone will read a field and ask why
+it exists, and the answer will name you.
+
 ## Versioning
 
 Semantic versioning on the package, independent of the `contract` number. Pin an **exact** version:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "discovery-media-player",
-  "version": "0.1.152",
+  "version": "0.1.154",
   "description": "Self-hosted document viewer: per-recipient tracked links, reading analytics, live presentation. The core knows nothing about the application hosting it — everything it borrows arrives through an injected context.",
   "keywords": [
     "pdf-viewer",

package/server/mesures.js

@@ -180,6 +180,22 @@ function relever() {
     // ⚠️ L'ÉCHELLE EST PUBLIÉE AVEC LES CHIFFRES. Sans elle, `p95sousMs: 250` ne dit pas si la
     // mesure suivante aurait pu être 251 ou 999 — un lecteur ne peut pas juger de sa précision.
     seauxMs: SEAUX_MS,
+    // ⚠️ LE DÉNOMINATEUR DE `routes`, ET IL MANQUAIT DEPUIS LE DÉBUT. Une famille sans échantillon
+    // est OMISE de `routes` — donc `routes: {}` ne distingue pas « aucun trafic sur la fenêtre »
+    // de « la mesure ne tourne pas ». Chez un hôte chargé la question ne se pose jamais : il y a
+    // toujours des entrées, et leur présence témoigne d'elle-même. Chez un hôte à 99 sessions,
+    // rien ne témoigne de rien.
+    //
+    // La règle vient d'un hôte, sur ses propres volumes : « les champs dont la valeur est son
+    // propre témoin à grande échelle ont besoin d'un témoin explicite à petite échelle ». Une
+    // petite installation n'est pas seulement privée d'occasions — elle perd du POUVOIR
+    // DISCRIMINANT, et un hôte plus chargé se trouve mieux instrumenté sans avoir rien instrumenté.
+    //
+    // ⚠️ ET LES DEUX CHAMPS VOISINS AVAIENT DÉJÀ RAISON, ce qui rend l'omission mesurable plutôt
+    // qu'opinable : `statuts` publie ses cinq clés à zéro, `boucleMs` publie `n: 0` avec des
+    // `null` explicites « plutôt que de publier un zéro qui se lirait la boucle est saine ». Trois
+    // champs frères du même objet, deux qui portent leur dénominateur et un qui l'oubliait.
+    familles: [...FAMILLES],
     routes,
     base: centiles(histoBase),
     statuts: { ...statuts },

package/server/retention.js

@@ -455,6 +455,16 @@ function tick() {
 // pas une opinion sur ce qu'un hôte peut avoir.
 const BORNE_RESTE = 5000;
 
+// ⚠️ DOUZE SECONDES, ET LE NOMBRE VIENT D'ÉVITER UNE ÉGALITÉ, PAS D'UN GOÛT. Il valait 8000 — très
+// exactement le `statement_timeout` qu'un hôte a mesuré sur son rôle `authenticator`. Deux minuteries
+// réglées sur la même valeur ne rendent pas un résultat faux ici (les deux voies retombent sur
+// `null`, un banc l'éprouve), mais elles rendent la CAUSE indécidable : quand notre abandon gagne la
+// course, le `57014` du serveur ne nous parvient jamais, et « la requête était trop lente » devient
+// indistinguable de « le réseau est tombé ». Un plafond client strictement AU-DESSUS des plafonds
+// serveur courants laisse le serveur expliquer son refus. C'est la règle de la marge sur le SUJET,
+// appliquée à du code de production plutôt qu'à un banc.
+const DELAI_LECTURE = 12000;
+
 const SONDES_RESTE = [
   ["sessionsIp", "commercial_doc_sessions", "session_id", "ip"],
   ["sessionsUa", "commercial_doc_sessions", "session_id", "ua"],
@@ -543,7 +553,7 @@ async function resteApres(chemin, cle, dernier) {
   if (dernier == null) return true;
   try {
     const suite = await PLAYER.db.request(
-      `${chemin}&${cle}=gt.${enc(String(dernier))}&order=${cle}.asc&limit=1`, { timeoutMs: 8000 });
+      `${chemin}&${cle}=gt.${enc(String(dernier))}&order=${cle}.asc&limit=1`, { timeoutMs: DELAI_LECTURE });
     // Pas de réponse analysable ⇒ on ne sait pas ⇒ « au moins ». Se tromper vers le minorant ne
     // fait que sous-estimer ; se tromper vers l'exactitude fait conclure.
     return !Array.isArray(suite) || suite.length > 0;
@@ -596,7 +606,7 @@ async function compterBorne(chemin, cle) {
     // ⚠️ ET L'ORDRE N'EST PAS DÉCORATIF : sans lui, « la dernière ligne reçue » ne désigne aucune
     // frontière, et le curseur de la sonde ne voudrait rien dire.
     const lignes = await PLAYER.db.request(
-      `${chemin}&order=${cle}.asc&limit=${BORNE_RESTE + 1}`, { timeoutMs: 8000 });
+      `${chemin}&order=${cle}.asc&limit=${BORNE_RESTE + 1}`, { timeoutMs: DELAI_LECTURE });
     if (!Array.isArray(lignes)) return compte(null, false, VOIE_BORNEE);
     // Notre propre borne atteinte : la preuve est dans la ligne excédentaire, rien à demander.
     if (lignes.length > BORNE_RESTE) return compte(BORNE_RESTE, true, VOIE_BORNEE);