Docstore documentation

kohler · kohler · commit a1caa74f9371 · 2024-09-27T14:53:26.000-04:00
diff --git a/batch/cleandocstore.php b/batch/cleandocstore.php
@@ -30,6 +30,8 @@ class CleanDocstore_Batch {
     public $dry_run;
     /** @var bool */
     public $keep_temp;
+    /** @var bool */
+    public $only_temp;
     /** @var int */
     public $cutoff;
     /** @var DocumentHashMatcher */
@@ -49,10 +51,17 @@ function __construct(Conf $conf, $docstores, $arg) {
         $this->verbose = isset($arg["verbose"]);
         $this->dry_run = isset($arg["dry-run"]);
         $this->keep_temp = isset($arg["keep-temp"]);
+        $this->only_temp = isset($arg["only-temp"]) || !$conf->s3_client();
         $this->cutoff = isset($arg["all"]) ? Conf::$now + 86400 : Conf::$now - 86400;
         $this->hash_matcher = new DocumentHashMatcher($arg["match"] ?? null);
     }
 
+    /** @param DocumentFileTreeMatch $fm
+     * @return bool */
+    static function is_temp($fm) {
+        return $fm->tree->treeid === 1;
+    }
+
     /** @return ?DocumentFileTreeMatch */
     function fparts_random_match() {
         $fmatches = [];
@@ -66,7 +75,7 @@ function fparts_random_match() {
                  ++$j) {
                 $fm = $ftree->random_match();
                 if ($fm->is_complete()
-                    && (($fm->treeid & 1) === 0
+                    && (!self::is_temp($fm)
                         || max($fm->atime(), $fm->mtime()) < $this->cutoff)) {
                     ++$n;
                     $fmatches[] = $fm;
@@ -75,7 +84,8 @@ function fparts_random_match() {
                 }
             }
             if ($n === 0) {
-                $this->ftrees[$i] = null;
+                array_splice($this->ftrees, $i, 1);
+                --$i;
             }
         }
         usort($fmatches, function ($a, $b) {
@@ -86,11 +96,11 @@ function fparts_random_match() {
                 return $at ? -1 : ($bt ? 1 : 0);
             }
             $aage = Conf::$now - $at;
-            if ($a->treeid & 1) {
+            if (self::is_temp($a)) {
                 $aage = $aage > 604800 ? 100000000 : $aage * 2;
             }
             $bage = Conf::$now - $bt;
-            if ($b->treeid & 1) {
+            if (self::is_temp($b)) {
                 $bage = $bage > 604800 ? 100000000 : $bage * 2;
             }
             return $bage <=> $aage;
@@ -99,7 +109,7 @@ function fparts_random_match() {
             return null;
         } else {
             $fm = $fmatches[0];
-            $this->ftrees[$fm->treeid]->hide($fm);
+            $fm->tree->hide($fm);
             return $fm;
         }
     }
@@ -142,7 +152,7 @@ function run() {
         }
 
         if (empty($this->docstores) || !$this->conf->docstore()) {
-            throw new ErrorException("No docstore to clean");
+            throw new CommandLineException("No docstore to clean");
         }
 
         preg_match('/\A((?:\/[^\/%]*(?=\/|\z))+)/', $this->docstores[0], $m);
@@ -153,7 +163,7 @@ function run() {
             $ts = disk_total_space($usage_directory);
             $fs = disk_free_space($usage_directory);
             if ($ts === false || $fs === false) {
-                throw new ErrorException("{$usage_directory}: Cannot evaluate free space");
+                throw new CommandLineException("{$usage_directory}: Cannot evaluate free space");
             } else if ($fs >= $ts * (1 - ($this->max_usage ?? $this->min_usage))) {
                 if (!$this->quiet) {
                     fwrite(STDOUT, "{$usage_directory}: free space sufficient\n");
@@ -170,13 +180,13 @@ function run() {
 
         foreach ($this->docstores as $i => $dp) {
             if (!str_starts_with($dp, "/") || strpos($dp, "%") === false) {
-                throw new ErrorException("{$dp}: Bad docstore pattern");
+                throw new CommandLineException("{$dp}: Bad docstore pattern");
+            }
+            if (!$this->only_temp) {
+                $this->ftrees[] = new DocumentFileTree($dp, $this->hash_matcher, 0);
             }
-            $this->ftrees[] = new DocumentFileTree($dp, $this->hash_matcher, count($this->ftrees));
             if (!$this->keep_temp) {
-                $this->ftrees[] = new DocumentFileTree(Filer::docstore_fixed_prefix($dp) . "tmp/%w", $this->hash_matcher, count($this->ftrees));
-            } else {
-                $this->ftrees[] = null;
+                $this->ftrees[] = new DocumentFileTree(Filer::docstore_fixed_prefix($dp) . "tmp/%w", $this->hash_matcher, 1);
             }
         }
 
@@ -185,8 +195,7 @@ function run() {
         while ($count > 0
                && ($usage_threshold === null || $bytesremoved < $usage_threshold)
                && ($fm = $this->fparts_random_match())) {
-            if (($fm->treeid & 1) !== 0
-                || $this->check_match($fm)) {
+            if (self::is_temp($fm) || $this->check_match($fm)) {
                 $size = filesize($fm->fname);
                 if ($this->dry_run || unlink($fm->fname)) {
                     if ($this->verbose) {
@@ -205,30 +214,31 @@ function run() {
         if (!$this->quiet) {
             fwrite(STDOUT, $usage_directory . ": " . ($this->dry_run ? "would remove " : "removed ") . plural($nsuccess, "file") . ", " . plural($bytesremoved, "byte") . "\n");
         }
-        if ($nsuccess == 0) {
-            fwrite(STDERR, "Nothing to delete\n");
+        if ($nsuccess === 0 && !$this->quiet) {
+            fwrite(STDERR, "Nothing to clean\n");
         }
-        return $nsuccess && $nsuccess == $ndone ? 0 : 1;
+        return $nsuccess > 0 && $nsuccess === $ndone ? 0 : 1;
     }
 
     /** @return CleanDocstore_Batch */
     static function make_args($argv) {
         $arg = (new Getopt)->long(
             "name:,n: !",
             "config: !",
-            "help,h",
             "count:,c: {n} =COUNT Clean up to COUNT files",
             "match:,m: =MATCH Clean files matching MATCH",
-            "verbose,V",
             "dry-run,d Do not remove files",
             "max-usage:,u: {f} =FRAC Clean until usage is below FRAC",
             "min-usage:,U: {f} =FRAC Do not clean if usage is below FRAC",
             "all Clean all files, including files recently modified",
-            "quiet,silent,q",
-            "keep-temp",
-            "docstore"
+            "quiet,silent,q Be quiet",
+            "keep-temp Keep temporary files",
+            "only-temp Only clean temporary files",
+            "help,h",
+            "verbose,V Be more verbose",
+            "docstore Output docstore patterns and exit"
         )->helpopt("help")
-         ->description("Remove files from HotCRP docstore that are on S3.
+         ->description("Remove old files from HotCRP docstore
 Usage: php batch/cleandocstore.php [-c COUNT|-u FRAC] [-V] [-d] [DOCSTORES...]\n")
          ->parse($argv);
 
diff --git a/devel/manual/docstore.md b/devel/manual/docstore.md
@@ -0,0 +1,66 @@
+# HotCRP document store
+
+HotCRP submission metadata are stored in the per-conference MySQL database. In
+default installations, document contents, such as PDF files, are also stored
+in the database. However, the system can also be configured to store document
+contents on the filesystem in a **document store** or **docstore**.
+
+
+## Configuring the docstore
+
+Set up a document store by configuring `$Opt["docstore"]`. This setting is a
+filename pattern that sets where documents are stored on the filesystem.
+
+To determine the filename for a document, HotCRP expands `%` escapes in
+`$Opt["docstore"]` using document information. The escapes are:
+
+| Escape | Meaning | Examples  |
+|:-------|:--------|:---------|
+| `%H`   | Content hash | `d16c7976d9081368c7dca2da3a771065c3222069a1ad80dcd99d972b2efadc8b` |
+| `%NH`  | First `N` bytes of content hash | `d16` (for `%3H`) |
+| `%a`   | Hash algorithm | `sha256`, `sha1` |
+| `%A`   | Hash algorithm prefix | `sha2-` (for SHA-256), empty string (for SHA-1) |
+| `%h`   | Content hash with algorithm prefix | `sha2-d16c7976d9081368c7dca2da3a771065c3222069a1ad80dcd99d972b2efadc8b` |
+| `%Nh`  | First `N` bytes of content hash with algorithm prefix | `sha2-d16` (for `%3h`) |
+| `%x`   | File extension | `.pdf`, `.txt` |
+| `%%`   | Literal `%` | `%` |
+
+A full `$Opt["docstore"]` setting must include a full hash (`%H` or `%h`). If
+`$Opt["docstore"]` does not include a `%` sign, then HotCRP automatically
+appends `/%h%x` to the setting value, and if `$Opt["docstore"]` is `true`,
+HotCRP uses `docs/%h%x`. Relative paths are interpreted relative to the HotCRP
+installation directory.
+
+The HotCRP PHP server must have read and write permission to the document
+store. `php-fpm` and/or `httpd` typically own the docstore directory, or they
+have group access (and the docstore direcrory has set-group-id permission).
+HotCRP will create subdirectories as necessary; for instance, with docstore
+`"docs/%2h/%H%x`, HotCRP might try to create the docstore subdirectory
+`docs/sha2-d1` to fit a file with SHA-256 hash
+`d16c7976d9081368c7dca2da3a771065c3222069a1ad80dcd99d972b2efadc8b`.
+
+
+## Temporary docstore
+
+A special subdirectory of the docstore is used for large temporary files,
+especially files that may need to outlive a single request. Examples include
+chunks of uploaded documents and constructed ZIP archives and CSV files.
+
+To form the temporary docstore, HotCRP appends `/tmp` to the docstore’s fixed
+prefix. For example, the docstore `"/home/hotcrp/docs/sub-%3h/%H%x"` has
+temporary docstore `/home/hotcrp/docs/tmp`.
+
+The temporary docstore should be cleaned periodically, for instance by the
+batch script `php batch/cleandocstore.php`.
+
+
+## Docstore, database, and S3
+
+HotCRP can store documents in the MySQL database (the default), in the
+docstore, and in Amazon S3. Amazon S3 is typically the slowest of these
+methods, but needs no separate backup. If you have configured either the
+docstore or S3, you can disable database storage by setting
+`$Opt["dbNoPapers"]` to `true`. If you have configured the docstore *and* S3,
+then the docstore can act as a cache for S3. Incoming documents are stored in
+both places; if a docstore version is missing later, HotCRP will check S3 for
+it.
diff --git a/devel/manual/index.md b/devel/manual/index.md
@@ -9,6 +9,7 @@ configuring features inaccessible through the public settings UI.
 * [**Components**](./components.md): How HotCRP configuration JSON files work
 * [**Message formatting and translation**](./fmt.md)
 * [**Page configuration**](./pages.md)
+* [**Document store**](./docstore.md)
 
 ## Specific pages
 
diff --git a/devel/manual/sessions.md b/devel/manual/sessions.md
@@ -1,14 +1,12 @@
-# HotCRP session data
+# HotCRP sessions
 
-This page describes the format of HotCRP session data.
+Each HotCRP session contains both **global** data, relevant for all
+conferences attached to a session, and **conference** data relevant to a
+single conference. Session data is stored in PHP’s native [session
+encoding](https://www.php.net/manual/en/function.session-encode.php).
 
-HotCRP sessions contain both **global** data, relevant for all conferences
-attached to a session, and **conference** data, which is relevant to a single
-conference. Conferences are distinguished by their **session keys**, which are
-`@` followed by the conference’s database name.
 
-
-## Global keys
+## Global session data
 
 ### Session version and expiration
 
@@ -87,7 +85,7 @@ conference. Conferences are distinguished by their **session keys**, which are
 * `login_bounce`
 
 
-## Conference keys
+## Conference session data
 
 Session data relevant to one conference is stored in the session element named
 by the conference’s session key, e.g., `@db-sigcomm23`. This element is
diff --git a/etc/distoptions.php b/etc/distoptions.php
@@ -115,15 +115,13 @@
 //   docstore        Set to true to serve papers and other downloads from a
 //                   cache on the local filesystem. By default this cache is
 //                   created in the "docs" directory. You can also set
-//                   $Opt["docstore"] to a directory name.
-//   docstoreSubdir  Set to true (or a small number, like 3) if the document
-//                   store should use subdirectories. This can be useful if
-//                   you expect thousands of submissions.
+//                   $Opt["docstore"] to a directory name, or to a directory
+//                   pattern such as "docs/%2H/%h%x".
 //   s3_bucket       Amazon S3 bucket name to store paper submissions.
 //   s3_key          Amazon AWS access key ID (used for S3).
 //   s3_secret       Amazon AWS secret access key (used for S3).
 //   dbNoPapers      Set to true to not store papers in the database.
-//                   Requires filestore, S3 storage, or both.
+//                   Requires docstore, S3 storage, or both.
 
 
 // TIMES AND DATES
diff --git a/src/api/api_tags.php b/src/api/api_tags.php
@@ -111,9 +111,7 @@ static function run(Contact $user, $qreq, $prow) {
         }
 
         $assigner = new AssignmentSet($user);
-        if ($prow) {
-            $assigner->enable_papers($prow);
-        }
+        $assigner->enable_papers($prow);
         $assigner->parse(join("\n", $x));
         $mlist = $assigner->message_list();
         $ok = $assigner->execute();
@@ -144,9 +142,8 @@ static function run(Contact $user, $qreq, $prow) {
     }
 
     /** @param Qrequest $qreq
-     * @param ?PaperInfo $prow
      * @return JsonResult */
-    static function assigntags(Contact $user, $qreq, $prow) {
+    static function assigntags(Contact $user, $qreq) {
         if (!isset($qreq->tagassignment)) {
             return JsonResult::make_missing_error("tagassignment");
         }
@@ -162,9 +159,6 @@ static function assigntags(Contact $user, $qreq, $prow) {
         }
 
         $assigner = new AssignmentSet($user);
-        if ($prow) {
-            $assigner->enable_papers($prow);
-        }
         $assigner->parse(join("\n", $x));
         $mlist = $assigner->message_list();
         $ok = $assigner->execute();
diff --git a/src/conference.php b/src/conference.php
@@ -674,13 +674,6 @@ function refresh_options() {
             $dpath = $docstore;
         } else if ($docstore === true) {
             $dpath = "docs";
-        } else if ($docstore === null && isset($this->opt["filestore"])) {
-            if (is_string($this->opt["filestore"])) {
-                $dpath = $this->opt["filestore"];
-            } else if ($this->opt["filestore"] === true) {
-                $dpath = "filestore";
-            }
-            $dpsubdir = $this->opt["filestoreSubdir"] ?? null;
         }
         if ($dpath !== "") {
             if ($dpath[0] !== "/") {
diff --git a/src/documentfiletree.php b/src/documentfiletree.php
@@ -277,7 +277,7 @@ static function random_index($di) {
     /** @return DocumentFileTreeMatch */
     function first_match(?DocumentFileTreeMatch $after = null) {
         $this->clear();
-        $fm = new DocumentFileTreeMatch($this->treeid);
+        $fm = new DocumentFileTreeMatch($this);
         for ($i = 0; $i < $this->_n; ++$i) {
             if ($i % 2 === 0) {
                 $fm->fname .= $this->_components[$i];
@@ -297,7 +297,7 @@ function first_match(?DocumentFileTreeMatch $after = null) {
     /** @return DocumentFileTreeMatch */
     function random_match() {
         $this->clear();
-        $fm = new DocumentFileTreeMatch($this->treeid);
+        $fm = new DocumentFileTreeMatch($this);
         for ($i = 0; $i < $this->_n; ++$i) {
             if ($i % 2 === 0) {
                 $fm->fname .= $this->_components[$i];
@@ -316,7 +316,7 @@ function random_match() {
 
     function hide(DocumentFileTreeMatch $fm) {
         // account for removal
-        assert($fm->treeid === $this->treeid);
+        assert($fm->tree === $this);
         for ($i = count($fm->idxes) - 1; $i >= 0; --$i) {
             $this->_dirinfo[$fm->bdirs[$i]]->hide_component_index($fm->idxes[$i]);
         }
@@ -342,8 +342,9 @@ function jsonSerialize() {
 }
 
 class DocumentFileTreeMatch {
-    /** @var int */
-    public $treeid;
+    /** @var DocumentFileTree
+     * @readonly */
+    public $tree;
     /** @var list<string> */
     public $bdirs = [];
     /** @var list<int> */
@@ -359,9 +360,9 @@ class DocumentFileTreeMatch {
     /** @var null|int|false */
     private $_mtime;
 
-    /** @param int $treeid */
-    function __construct($treeid) {
-        $this->treeid = $treeid;
+    /** @param DocumentFileTree $tree */
+    function __construct($tree) {
+        $this->tree = $tree;
     }
     /** @param int $idx
      * @param string $suffix */
