Build your personal knowledge base with Trilium Notes
TypeScript electronelectron-appknowledge-baseknowledge-graphknowledge-managementknowledge-management-graphlocal-firstnote-managmentnote-takernote-takingnotebooknotesnotes-apppersonal-knowledge-basepersonal-wikiscriptableself-hostedself-hostingwiki
chore(mobile): remove the save and backup timing logs
Elian Doran1ee3c376479c
diff --git a/.claude/skills/developing-capacitor-mobile/SKILL.md b/.claude/skills/developing-capacitor-mobile/SKILL.mdindex 79a24e1653..81de59f1d6 100644--- a/.claude/skills/developing-capacitor-mobile/SKILL.md+++ b/.claude/skills/developing-capacitor-mobile/SKILL.md@@ -93,11 +93,11 @@ sheet.`capacitor.config.json` (iOS only builds what is listed), and `cap update android` to regenerate thetracked `capacitor.settings.gradle` / `app/capacitor.build.gradle`. iOS's `CapApp-SPM/Package.swift`is gitignored and regenerated by CI.-- **Throughput is measured, not guessed.** Every save logs one `[Save]` line (total, MB/s, encode vs- transport-wait split, transport used) and the worker logs a `[Backup]` line for backups (reads vs- waiting-on-consumer vs hashing/crypto). Read those before optimizing: the remaining backup-side- cost on a phone is the container's incremental SHA-256, which is `@noble/hashes` pure JS- (`packages/trilium-backup-container/src/backend-web.ts`) because WebCrypto has no streaming digest.+- **The known cost profile** (1.8 GB backup on a mid-range phone, Sept 2026): ~16s, split evenly+ between SQLite page reads and the container's WASM SHA-256 (`hash-wasm` in+ `packages/trilium-backup-container/src/backend-web.ts`); the sink's writes are not a factor. The+ base64 plugin bridge caps around 13 MB/s (`MessageHandler` re-parses each call's JSON whole with+ `org.json`), which is why the sink exists.- **The database backup takes the same last step, from a different source.** `saveDatabase()` in`local-bridge.ts` consumes the worker's `BACKUP_STREAM` channel in the page rather than relaying theport to the service worker, so no SW is involved (which is also why it works on iOS), anddiff --git a/apps/standalone/src/lightweight/backup-download.ts b/apps/standalone/src/lightweight/backup-download.tsindex e95683c9fa..1d103f0dd3 100644--- a/apps/standalone/src/lightweight/backup-download.ts+++ b/apps/standalone/src/lightweight/backup-download.ts@@ -6,8 +6,7 @@ import {import {type LiveDatabaseReader,- streamLiveDatabasePages,- type StreamTiming+ streamLiveDatabasePages} from "./backup-stream.js";/**@@ -130,23 +129,13 @@ export async function streamDatabaseDownload(return requests.shift() ?? "cancel";};- /**- * How long was spent with a chunk ready and nowhere to put it.- *- * The one thing this side cannot make faster: the browser asks for the next piece as it commits- * the last one to disk, so a large figure here means the download itself is the limit and no- * amount of reading or hashing faster would show up in the total.- */- let waitedMs = 0;let sentBytes = 0;let totalBytes = 0;let reportedAt = 0;/** Posts one chunk per granted pull, or throws the stop that was granted instead. */const sendChunk = async (chunk: Uint8Array): Promise<void> => {- const waitingSince = Date.now();const request = await nextRequest();- waitedMs += Date.now() - waitingSince;if (request !== "pull") {throw new DownloadStopped(request);@@ -169,9 +158,8 @@ export async function streamDatabaseDownload(};let source: ReadableStream<Uint8Array> | undefined;- const startedAt = Date.now();try {- const { byteSize, stream, timing } = streamLiveDatabasePages(db);+ const { byteSize, stream } = streamLiveDatabasePages(db);source = stream;// A container either way, so every backup this application writes has one shape and one@@ -196,7 +184,6 @@ export async function streamDatabaseDownload(// The last one is unconditional: a throttled report would otherwise leave the screen a// chunk short of the total it is about to be told is finished.options.onProgress?.(sentBytes, totalBytes);- reportTiming(byteSize, timing, waitedMs, Date.now() - startedAt);return { status: "done" };} catch (e) {@@ -220,34 +207,6 @@ export async function streamDatabaseDownload(}}-/**- * One line saying where a backup's time went, split three ways so a slow one can be acted on.- *- * `read` is the database coming out through SQLite, `waiting` is having a chunk ready and the- * download not yet asking for it, and what neither accounts for is the hashing and encryption in- * between. Only the first is worth optimising: waiting is the browser committing the file to disk,- * which is the floor, and the rest is a hash the format cannot do without.- */-function reportTiming(- byteSize: number,- timing: StreamTiming,- waitedMs: number,- totalMs: number-): void {- const mib = byteSize / (1024 * 1024);- const rate = (ms: number) => (ms > 0 ? mib / (ms / 1000) : 0).toFixed(1);- const share = (ms: number) => Math.round(100 * ms / Math.max(totalMs, 1));- const restMs = Math.max(0, totalMs - timing.readMs - waitedMs);-- console.log(- `[Backup] ${mib.toFixed(0)} MiB in ${(totalMs / 1000).toFixed(1)}s (${rate(totalMs)} MiB/s). `- + `Reads ${(timing.readMs / 1000).toFixed(1)}s (${share(timing.readMs)}%, `- + `${rate(timing.readMs)} MiB/s over ${timing.pages} pages), `- + `waiting on the download ${(waitedMs / 1000).toFixed(1)}s (${share(waitedMs)}%), `- + `hashing and crypto ${(restMs / 1000).toFixed(1)}s (${share(restMs)}%).`- );-}-/** A best-effort message: the port may already be gone, and this report is all that would say so. */function post(port: DownloadPort, message: unknown): void {try {diff --git a/apps/standalone/src/lightweight/backup-stream.ts b/apps/standalone/src/lightweight/backup-stream.tsindex 2e8f9ad3eb..4df1416059 100644--- a/apps/standalone/src/lightweight/backup-stream.ts+++ b/apps/standalone/src/lightweight/backup-stream.ts@@ -32,14 +32,6 @@ export interface LiveDatabaseReader {getColumn(sql: string, params?: unknown[]): unknown[];}-/** What the read cost, for a caller that wants to say where a slow backup spent its time. */-export interface StreamTiming {- /** Milliseconds spent inside the page reads, as against waiting on whatever consumes them. */- readMs: number;- /** Pages handed over so far. */- pages: number;-}-/** Thrown by {@link streamLiveDatabasePages} when a write lands mid-stream. The caller retries. */export class DatabaseChangedError extends Error {constructor() {@@ -52,8 +44,6 @@ export interface DatabaseStream {/** Exact size of the streamed database, known before the first byte flows. */byteSize: number;stream: ReadableStream<Uint8Array>;- /** Updated as the stream runs, so a caller can report where the time went. */- timing: StreamTiming;}/**@@ -75,15 +65,12 @@ export function streamLiveDatabasePages(db: LiveDatabaseReader): DatabaseStreamconst pageCount = countOf(db.getValue("PRAGMA page_count"), "page_count");const fingerprint = fingerprintOf(db);- const timing: StreamTiming = { readMs: 0, pages: 0 };let nextPage = 1;return {byteSize: pageSize * pageCount,- timing,stream: new ReadableStream<Uint8Array>({pull(controller) {- const startedAt = Date.now();try {if (fingerprintOf(db) !== fingerprint) {throw new DatabaseChangedError();@@ -96,13 +83,10 @@ export function streamLiveDatabasePages(db: LiveDatabaseReader): DatabaseStreamconst pages = Math.min(BATCH_PAGES, pageCount - nextPage + 1);const batch = readLivePages(db, nextPage, pages, pageSize);nextPage += pages;- timing.pages += pages;controller.enqueue(batch);} catch (e) {controller.error(e);- } finally {- timing.readMs += Date.now() - startedAt;}}})diff --git a/apps/standalone/src/services/capacitor_download.ts b/apps/standalone/src/services/capacitor_download.tsindex 7803831094..b31a6bbafb 100644--- a/apps/standalone/src/services/capacitor_download.ts+++ b/apps/standalone/src/services/capacitor_download.ts@@ -109,11 +109,7 @@ export async function saveChunksToDevice(target: SaveTarget): Promise<StandaloneSaveResult> {try {- const started = performance.now();- const outcome = await writeChunks(fileName, chunks, target);- reportThroughput(fileName, outcome, performance.now() - started);- const { uri } = outcome;-+ const uri = await writeChunks(fileName, chunks, target);return { ...await shareFile(uri, fileName), location: displayPath(uri) };} catch (e) {return {@@ -161,20 +157,12 @@ function getPlugin<T>(name: string): T {return plugin as T;}-interface WriteOutcome {- uri: string;- bytes: number;- encodeMs: number;- waitMs: number;- transport: "sink" | "plugin";-}-/** Writes the chunks into `target` one at a time and returns the finished file's URI. */async function writeChunks(fileName: string,chunks: AsyncIterable<Uint8Array>,{ directory, folder, sweep }: SaveTarget-): Promise<WriteOutcome> {+): Promise<string> {const filesystem = getPlugin<FilesystemPlugin>("Filesystem");const path = `${folder}/${fileName}`;@@ -192,15 +180,13 @@ async function writeChunks(await filesystem.writeFile({ path, data: "", directory, recursive: true });const { uri } = await filesystem.getUri({ path, directory });- return { uri, transport: "sink", ...await writeThroughSink(sink, displayPath(uri), chunks) };+ await writeThroughSink(sink, displayPath(uri), chunks);+ return uri;} finally {releaseFileSink();}}- let bytes = 0;- let encodeMs = 0;- let waitMs = 0;let started = false;// One call is kept in flight while the next chunk encodes, so the encode's cost hides behind// the bridge's. Never two: appends are ordered only because each is issued after the previous@@ -209,25 +195,20 @@ async function writeChunks(const awaitPending = async () => {if (pending) {- const waitStart = performance.now();await pending;- waitMs += performance.now() - waitStart;pending = null;}};await withoutCallLogging(async () => {for await (const chunk of rechunk(chunks, chunkBytes)) {- const encodeStart = performance.now();const data = toBase64(chunk);- encodeMs += performance.now() - encodeStart;await awaitPending();pending = started? filesystem.appendFile({ path, data, directory }): filesystem.writeFile({ path, data, directory, recursive: true });started = true;- bytes += chunk.length;}await awaitPending();});@@ -237,7 +218,7 @@ async function writeChunks(}const { uri } = await filesystem.getUri({ path, directory });- return { uri, bytes, encodeMs, waitMs, transport: "plugin" };+ return uri;}/**@@ -281,10 +262,7 @@ async function writeThroughSink(sink: NativeFileSink,filePath: string,chunks: AsyncIterable<Uint8Array>-): Promise<{ bytes: number; encodeMs: number; waitMs: number }> {- let bytes = 0;- let waitMs = 0;-+): Promise<void> {// Replies arrive in send order, so the head of this queue is always the awaited step.const waiting: ((reply: string) => void)[] = [];const listener = (event: { data: unknown }) => waiting.shift()?.(String(event.data));@@ -312,17 +290,12 @@ async function writeThroughSink(try {await send(JSON.stringify({ type: "open", path: filePath }), "opened");for await (const chunk of rechunk(chunks, chunkBytes)) {- const waitStart = performance.now();await send(exactBuffer(chunk), "written");- waitMs += performance.now() - waitStart;- bytes += chunk.length;}await send(JSON.stringify({ type: "close" }), "closed");} finally {sink.removeEventListener("message", listener);}-- return { bytes, encodeMs: 0, waitMs };}/** The chunk's bytes as exactly one ArrayBuffer, which is what the message channel carries. */@@ -359,32 +332,6 @@ async function withoutCallLogging<T>(write: () => Promise<T>): Promise<T> {}}-/**- * What the write achieved, which is the only place the transport's cost per byte is visible.- *- * The split names the suspect: `base64` is time in {@link toBase64}, the wait is time blocked on- * the transport. What is left of the total is producing the bytes (the worker, for a backup) plus- * everything the pipelining managed to hide.- */-function reportThroughput(- fileName: string,- { bytes, encodeMs, waitMs, transport }: WriteOutcome,- elapsedMs: number-): void {- const seconds = elapsedMs / 1000;- const megabytes = bytes / 1_000_000;- const rate = seconds > 0 ? (megabytes / seconds).toFixed(1) : "—";- const split = transport === "sink"- ? `via the binary sink — write wait ${(waitMs / 1000).toFixed(1)}s`- : `via the plugin bridge — base64 ${(encodeMs / 1000).toFixed(1)}s, `- + `bridge wait ${(waitMs / 1000).toFixed(1)}s`;-- console.log(- `[Save] ${fileName}: ${megabytes.toFixed(1)} MB in ${seconds.toFixed(1)}s (${rate} MB/s) `- + `${split}, ${Math.ceil(bytes / chunkBytes)} writes of ${chunkBytes / 1024} KiB`- );-}-/*** The file's URI as somewhere a person can be told to look. `getUri` answers with a `file://` URL,* percent-encoded, which is a path with two layers of machinery on top of it.