Run the backup job at every cron run; skip deflate on compressed data

lib/Repository/Repository.php

@@ -28,6 +28,10 @@ final class Repository {
 	public const FORMAT = 1;
 	public const BLOB_SIZE = 4 * 1048576;
 	private const FLAG_DEFLATE = 1;
+	/** Blobs from this size on get a quick compression test before the full compression. */
+	private const DEFLATE_PROBE_FROM = 256 * 1024;
+	/** Bytes of each of the three slices (start, middle, end) in that test. */
+	private const DEFLATE_PROBE_SLICE = 32 * 1024;
 	/** Reads per blob in a deep verify before it is reported as unreadable (not corrupted). */
 	private const VERIFY_READ_ATTEMPTS = 3;
 	/** Largest read that loadBlobs() makes of blobs that follow each other in a pack. */
@@ -703,7 +707,7 @@ final class Repository {
 		}
 		$flags = 0;
 		$payload = $data;
-		$deflated = gzdeflate($data, 6);
+		$deflated = self::worthDeflating($data) ? gzdeflate($data, 6) : false;
 		if ($deflated !== false && strlen($deflated) < strlen($data) * 0.95) {
 			$payload = $deflated;
 			$flags |= self::FLAG_DEFLATE;
@@ -713,6 +717,23 @@ final class Repository {
 		return $id;
 	}
 
+	/**
+	 * Photos, video and archives are compressed already: deflating them takes most of a backup's
+	 * CPU time (about 24 MiB/s) only to be thrown away, as it saves less than the 5% that keeps
+	 * it. A fast deflate of three small slices tells first; below 10% savings there, the full
+	 * deflate is skipped.
+	 */
+	private static function worthDeflating(string $data): bool {
+		$n = strlen($data);
+		if ($n < self::DEFLATE_PROBE_FROM) {
+			return true;
+		}
+		$k = self::DEFLATE_PROBE_SLICE;
+		$probe = substr($data, 0, $k) . substr($data, intdiv($n - $k, 2), $k) . substr($data, $n - $k);
+		$deflated = gzdeflate($probe, 1);
+		return $deflated !== false && strlen($deflated) * 10 < strlen($probe) * 9;
+	}
+
 	/**
 	 * Blobs spread over many packs (the tree segments of a run, one per step), through
 	 * IBackend::getRanges(): on a remote location one at a time is a round trip each, which for