Full restore over SFTP: read packs ahead on four connections at once

A restore read one pack at a time over one SFTP connection, which over a long line is bound by the round trip and by what one TCP connection gets: about 2-3 MiB/s from a Hetzner Storage Box to Canada, with PHP mostly waiting. Repository::restore() now first lists the ranges of packs the restore needs, in order (ReadAheadPlan), and the location reads them ahead into unlinked files in a spool folder in the target (IReadsAhead). Over SFTP, SftpRangeFetcher does that on four connections of its own, 16 requests in flight each, taking the answers of whichever connection has one. A range that fails, a damaged copy, or a blob that moved since the plan is read the usual way, which reports what is wrong. Other locations restore as before. On a test line with 120 ms RTT and 0.2% loss, a 1.6 GiB restore went from 5.1 to 17.8 MiB/s, all files identical.
2026-10-19 01:40:49 +02:00
parent b789d17636
commit ccbbf01c54
12 changed files with 1228 additions and 5 deletions
@@ -0,0 +1,123 @@
+<?php
+
+declare(strict_types=1);
+
+// SPDX-FileCopyrightText: 2026 Dillard Blom
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+namespace OCA\NgBackup\Repository;
+
+/**
+ * The ranges of packs a restore reads, in order, for IReadsAhead::readAhead(), and the reading of
+ * blobs from them as the restore goes.
+ *
+ * add() takes the blobs in the order the restore needs them. A blob in the same pack as the range
+ * before it, not before that range and at most GAP after its end, extends that range; anything
+ * else starts a new one. read() takes the same blobs in the same order: a blob is in the current
+ * range or in the next one, which then becomes current; a blob not in the plan is read elsewhere.
+ */
+final class ReadAheadPlan {
+	/** Unused bytes between two blobs of a pack that are read along rather than asked for apart. */
+	public const GAP = 4 * 1048576;
+	/** Longest range: packs are 32 MiB, repacked ones too. */
+	public const MAX_RANGE = 64 * 1048576;
+	/** Ranges after the current one that read() looks in for a blob. */
+	private const SEARCH = 16;
+
+	/** @var list<array{0:string, 1:int, 2:int}> [path, offset, length] */
+	private array $ranges = [];
+	/** @var \Generator<int, resource|null>|null */
+	private ?\Generator $streams = null;
+	private int $current = -1;
+	/** @var resource|null */
+	private $stream = null;
+
+	/** @param string $spool local folder for the ranges read ahead (see IReadsAhead::readAhead()) */
+	public function __construct(
+		public readonly string $spool,
+	) {
+	}
+
+	public function add(string $path, int $offset, int $length): void {
+		$last = array_pop($this->ranges);
+		if ($last !== null) {
+			[$p, $start, $len] = $last;
+			if ($p === $path && $offset >= $start && $offset <= $start + $len + self::GAP
+				&& $offset + $length - $start <= self::MAX_RANGE) {
+				$this->ranges[] = [$p, $start, max($len, $offset + $length - $start)];
+				return;
+			}
+			$this->ranges[] = $last;
+		}
+		$this->ranges[] = [$path, $offset, $length];
+	}
+
+	/** @return list<array{0:string, 1:int, 2:int}> */
+	public function ranges(): array {
+		return $this->ranges;
+	}
+
+	/** @param \Generator<int, resource|null> $streams what IReadsAhead::readAhead() gave for ranges() */
+	public function readFrom(\Generator $streams): void {
+		$this->streams = $streams;
+	}
+
+	/** The bytes of a blob, or null when its range could not be read ahead (read it the usual way). */
+	public function read(string $path, int $offset, int $length): ?string {
+		if ($this->streams === null) {
+			return null;
+		}
+		if (!$this->contains($this->current, $path, $offset, $length)) {
+			// Normally the next range. Further on when blobs before it were read elsewhere: the
+			// index was reloaded after another server's cleanup moved them to new packs.
+			$to = null;
+			for ($i = $this->current + 1; $i <= $this->current + self::SEARCH && $to === null; $i++) {
+				if ($this->contains($i, $path, $offset, $length)) {
+					$to = $i;
+				}
+			}
+			if ($to === null) {
+				return null; // not where the plan has it
+			}
+			if ($this->current < 0) {
+				$this->streams->current();
+				$this->current = 0;
+			}
+			while ($this->current < $to) {
+				$this->streams->next();
+				$this->current++;
+			}
+			$stream = $this->streams->valid() ? $this->streams->current() : null;
+			$this->stream = is_resource($stream) ? $stream : null;
+		}
+		$fh = $this->stream;
+		if ($fh === null || fseek($fh, $offset - $this->ranges[$this->current][1]) !== 0) {
+			return null;
+		}
+		$data = '';
+		while (strlen($data) < $length) {
+			$chunk = fread($fh, $length - strlen($data));
+			if ($chunk === false || $chunk === '') {
+				return null;
+			}
+			$data .= $chunk;
+		}
+		return $data;
+	}
+
+	/** Ends the reads ahead, also halfway: closes their streams and connections. */
+	public function close(): void {
+		$this->stream = null;
+		$streams = $this->streams;
+		$this->streams = null;
+		unset($streams); // the generator's finally blocks run when it is destroyed
+	}
+
+	private function contains(int $i, string $path, int $offset, int $length): bool {
+		if ($i < 0 || !isset($this->ranges[$i])) {
+			return false;
+		}
+		[$p, $start, $len] = $this->ranges[$i];
+		return $p === $path && $offset >= $start && $offset + $length <= $start + $len;
+	}
+}