ovsdb: Implement garbage collection.

debian/openvswitch-switch.init

@@ -232,6 +232,18 @@ case "$1" in
             cksum=`ovsdb-tool db-cksum "$conf_file" | awk '{print $1}'`
             cp "$conf_file" "$conf_file.backup$version-$cksum"
 
+            # Compact database.  This is important if the old schema did not
+            # enable garbage collection (i.e. if it did not have any tables
+            # with "isRoot": true) but the new schema does.  In that situation
+            # the old database may contain a transaction that creates a record
+            # followed by a transaction that creates the first use of the
+            # record.  Replaying that series of transactions against the new
+            # database schema (as "convert" does) would cause the record to be
+            # dropped by the first transaction, then the second transaction
+            # would cause a referential integrity failure (for a strong
+            # reference).
+            ovsdb-tool -vANY:console:emer compact $conf_file
+
             # Upgrade or downgrade schema and compact database.
             ovsdb-tool -vANY:console:emer convert $conf_file $schema_file
         fi

lib/ovsdb-data.c

@@ -285,9 +285,28 @@ parse_json_pair(const struct json *json,
     return NULL;
 }
 
+static void
+ovsdb_symbol_referenced(struct ovsdb_symbol *symbol,
+                        const struct ovsdb_base_type *base)
+{
+    assert(base->type == OVSDB_TYPE_UUID);
+
+    if (base->u.uuid.refTableName) {
+        switch (base->u.uuid.refType) {
+        case OVSDB_REF_STRONG:
+            symbol->strong_ref = true;
+            break;
+        case OVSDB_REF_WEAK:
+            symbol->weak_ref = true;
+            break;
+        }
+    }
+}
+
 static struct ovsdb_error * WARN_UNUSED_RESULT
 ovsdb_atom_parse_uuid(struct uuid *uuid, const struct json *json,
-                      struct ovsdb_symbol_table *symtab)
+                      struct ovsdb_symbol_table *symtab,
+                      const struct ovsdb_base_type *base)
 {
     struct ovsdb_error *error0;
     const struct json *value;
@@ -304,14 +323,17 @@ ovsdb_atom_parse_uuid(struct uuid *uuid, const struct json *json,
 
         error1 = unwrap_json(json, "named-uuid", JSON_STRING, &value);
         if (!error1) {
-            const char *name = json_string(value);
+            struct ovsdb_symbol *symbol;
 
             ovsdb_error_destroy(error0);
-            *uuid = ovsdb_symbol_table_insert(symtab, name)->uuid;
             if (!ovsdb_parser_is_id(json_string(value))) {
                 return ovsdb_syntax_error(json, NULL, "named-uuid string is "
                                           "not a valid <id>");
             }
+
+            symbol = ovsdb_symbol_table_insert(symtab, json_string(value));
+            *uuid = symbol->uuid;
+            ovsdb_symbol_referenced(symbol, base);
             return NULL;
         }
         ovsdb_error_destroy(error1);
@@ -321,10 +343,13 @@ ovsdb_atom_parse_uuid(struct uuid *uuid, const struct json *json,
 }
 
 static struct ovsdb_error * WARN_UNUSED_RESULT
-ovsdb_atom_from_json__(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
+ovsdb_atom_from_json__(union ovsdb_atom *atom,
+                       const struct ovsdb_base_type *base,
                        const struct json *json,
                        struct ovsdb_symbol_table *symtab)
 {
+    enum ovsdb_atomic_type type = base->type;
+
     switch (type) {
     case OVSDB_TYPE_VOID:
         NOT_REACHED();
@@ -364,7 +389,7 @@ ovsdb_atom_from_json__(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
         break;
 
     case OVSDB_TYPE_UUID:
-        return ovsdb_atom_parse_uuid(&atom->uuid, json, symtab);
+        return ovsdb_atom_parse_uuid(&atom->uuid, json, symtab, base);
 
     case OVSDB_N_TYPES:
     default:
@@ -384,7 +409,9 @@ ovsdb_atom_from_json__(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
  *
  * If 'symtab' is nonnull, then named UUIDs in 'symtab' are accepted.  Refer to
  * ovsdb/SPECS for information about this, and for the syntax that this
- * function accepts. */
+ * function accepts.  If 'base' is a reference and a symbol is parsed, then the
+ * symbol's 'strong_ref' or 'weak_ref' member is set to true, as
+ * appropriate. */
 struct ovsdb_error *
 ovsdb_atom_from_json(union ovsdb_atom *atom,
                      const struct ovsdb_base_type *base,
@@ -393,7 +420,7 @@ ovsdb_atom_from_json(union ovsdb_atom *atom,
 {
     struct ovsdb_error *error;
 
-    error = ovsdb_atom_from_json__(atom, base->type, json, symtab);
+    error = ovsdb_atom_from_json__(atom, base, json, symtab);
     if (error) {
         return error;
     }
@@ -440,9 +467,12 @@ ovsdb_atom_to_json(const union ovsdb_atom *atom, enum ovsdb_atomic_type type)
 }
 
 static char *
-ovsdb_atom_from_string__(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
-                         const char *s, struct ovsdb_symbol_table *symtab)
+ovsdb_atom_from_string__(union ovsdb_atom *atom,
+                         const struct ovsdb_base_type *base, const char *s,
+                         struct ovsdb_symbol_table *symtab)
 {
+    enum ovsdb_atomic_type type = base->type;
+
     switch (type) {
     case OVSDB_TYPE_VOID:
         NOT_REACHED();
@@ -503,7 +533,9 @@ ovsdb_atom_from_string__(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
 
     case OVSDB_TYPE_UUID:
         if (*s == '@') {
-            atom->uuid = ovsdb_symbol_table_insert(symtab, s)->uuid;
+            struct ovsdb_symbol *symbol = ovsdb_symbol_table_insert(symtab, s);
+            atom->uuid = symbol->uuid;
+            ovsdb_symbol_referenced(symbol, base);
         } else if (!uuid_from_string(&atom->uuid, s)) {
             return xasprintf("\"%s\" is not a valid UUID", s);
         }
@@ -535,7 +567,9 @@ ovsdb_atom_from_string__(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
  *        then an identifier beginning with '@' is also acceptable.  If the
  *        named identifier is already in 'symtab', then the associated UUID is
  *        used; otherwise, a new, random UUID is used and added to the symbol
- *        table.
+ *        table.  If 'base' is a reference and a symbol is parsed, then the
+ *        symbol's 'strong_ref' or 'weak_ref' member is set to true, as
+ *        appropriate.
  *
  * Returns a null pointer if successful, otherwise an error message describing
  * the problem.  On failure, the contents of 'atom' are indeterminate.  The
@@ -549,7 +583,7 @@ ovsdb_atom_from_string(union ovsdb_atom *atom,
     struct ovsdb_error *error;
     char *msg;
 
-    msg = ovsdb_atom_from_string__(atom, base->type, s, symtab);
+    msg = ovsdb_atom_from_string__(atom, base, s, symtab);
     if (msg) {
         return msg;
     }
@@ -1829,6 +1863,8 @@ ovsdb_symbol_table_put(struct ovsdb_symbol_table *symtab, const char *name,
     symbol = xmalloc(sizeof *symbol);
     symbol->uuid = *uuid;
     symbol->created = created;
+    symbol->strong_ref = false;
+    symbol->weak_ref = false;
     shash_add(&symtab->sh, name, symbol);
     return symbol;
 }

lib/ovsdb-data.h

@@ -236,6 +236,8 @@ struct ovsdb_symbol_table {
 struct ovsdb_symbol {
     struct uuid uuid;           /* The UUID that the symbol represents. */
     bool created;               /* Already used to create row? */
+    bool strong_ref;            /* Parsed a strong reference to this row? */
+    bool weak_ref;              /* Parsed a weak reference to this row? */
 };
 
 struct ovsdb_symbol_table *ovsdb_symbol_table_create(void);

lib/ovsdb-idl-provider.h

@@ -1,4 +1,4 @@
-/* Copyright (c) 2009, 2010 Nicira Networks.
+/* Copyright (c) 2009, 2010, 2011 Nicira Networks.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -47,6 +47,7 @@ struct ovsdb_idl_column {
 
 struct ovsdb_idl_table_class {
     char *name;
+    bool is_root;
     const struct ovsdb_idl_column *columns;
     size_t n_columns;
     size_t allocation_size;

ovsdb/SPECS

@@ -131,6 +131,7 @@ is represented by <database-schema>, as described below.
 
         "columns": {<id>: <column-schema>, ...}   required
         "maxRows": <integer>                      optional
+        "isRoot": <boolean>                       optional
 
     The value of "columns" is a JSON object whose names are column
     names and whose values are <column-schema>s.
@@ -152,12 +153,28 @@ is represented by <database-schema>, as described below.
         the database process is stopped and then started again, each
         "_version" also changes to a new random value.
 
+    If "isRoot" is omitted or specified as false, then any given row
+    in the table may exist only when there is at least one reference
+    to it, with refType "strong", from a different row (in the same
+    table or a different table).  This is a "deferred" action:
+    unreferenced rows in the table are deleted just before transaction
+    commit.  If "isRoot" is specified as true, then rows in the table
+    exist independent of any references (they can be thought of as
+    part of the "root set" in a garbage collector).
+
+    For compatibility with schemas created before "isRoot" was
+    introduced, if "isRoot" is omitted or false in every
+    <table-schema> in a given <database-schema>, then every table is
+    part of the root set.
+
     If "maxRows" is specified, as a positive integer, it limits the
     maximum number of rows that may be present in the table.  This is
     a "deferred" constraint, enforced only at transaction commit time
     (see the "transact" request below).  If "maxRows" is not
     specified, the size of the table is limited only by the resources
-    available to the database server.
+    available to the database server.  "maxRows" constraints are
+    enforced after unreferenced rows are deleted from tables with a
+    false "isRoot".
 
 <column-schema>
 

ovsdb/dot2pic

@@ -1,6 +1,6 @@
 #! /usr/bin/perl
 
-# Copyright (c) 2009, 2010 Nicira Networks
+# Copyright (c) 2009, 2010, 2011 Nicira Networks
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -29,7 +29,14 @@ while (<>) {
         $y *= $scale;
         $width *= $scale;
         $height *= $scale;
+        print "linethick = ", ($style eq 'bold' ? 0.5 : 1.0), ";\n";
         print "box at $x,$y wid $width height $height \"$name\"\n";
+        if ($style eq 'bold') {
+            my $inset = 2.0 / 72.0;
+            $width -= $inset * 2;
+            $height -= $inset * 2;
+            print "box at $x,$y wid $width height $height\n";
+        }
     } elsif (/edge/) {
         my (undef, $tail, $head, $n, $rest) = split(' ', $_, 5);
         my @xy;
@@ -51,6 +58,8 @@ while (<>) {
         }
         my ($style, $color) = split(' ', $rest);
 
+        print "linethick = ", ($style eq 'dotted' ? 0.5 : 1), ";\n";
+
         print "spline -> from $xy[0][0],$xy[0][1]";
         for (my ($i) = 0; $i <= $#xy; $i++) {
             print " to $xy[$i][0],$xy[$i][1]";

ovsdb/ovsdb-doc.in

@@ -290,9 +290,11 @@ Table	Purpose
 .SH "TABLE RELATIONSHIPS"
 .PP
 The following diagram shows the relationship among tables in the
-database.  Each node represents a table.  Each edge leads from the
+database.  Each node represents a table.  Tables that are part of the
+``root set'' are shown with double borders.  Each edge leads from the
 table that contains it and points to the table that its value
-represents.  Edges are labeled with their column names.
+represents.  Edges are labeled with their column names.  Thick lines
+represent strong references; thin lines represent weak references.
 .RS -1in
 """
         erStream = open(erFile, "r")

ovsdb/ovsdb-dot.in

@@ -16,6 +16,7 @@ def printEdge(tableName, baseType, label):
         options['label'] = '"%s"' % label
         if baseType.ref_type == 'weak':
             options['constraint'] = 'false'
+            options['style'] = 'dotted'
         print "\t%s -> %s [%s];" % (
             tableName,
             baseType.ref_table,
@@ -25,12 +26,17 @@ def schemaToDot(schemaFile):
     schema = ovs.db.schema.DbSchema.from_json(ovs.json.from_file(schemaFile))
 
     print "digraph %s {" % schema.name
+    print '\tsize="6.5,4";'
+    print '\tmargin="0";'
+    print "\tnode [shape=box];"
+    print "\tedge [dir=none, arrowhead=none, arrowtail=none];"
     for tableName, table in schema.tables.iteritems():
-        print '\tsize="6.5,4";'
-        print '\tmargin="0";'
-        print "\tnode [shape=box];"
-        print "\tedge [dir=none, arrowhead=none, arrowtail=none];"
-        print "\t%s;" % tableName
+        options = {}
+        if table.is_root:
+            options['style'] = 'bold'
+        print "\t%s [%s];" % (
+            tableName,
+            ', '.join(['%s=%s' % (k,v) for k,v in options.items()]))
         for columnName, column in table.columns.iteritems():
             if column.type.value:
                 printEdge(tableName, column.type.key, "%s key" % columnName)

ovsdb/ovsdb-idlc.in

@@ -487,7 +487,11 @@ static void\n%s_columns_init(void)
     print "struct ovsdb_idl_table_class %stable_classes[%sN_TABLES] = {" % (prefix, prefix.upper())
     for tableName, table in sorted(schema.tables.iteritems()):
         structName = "%s%s" % (prefix, tableName.lower())
-        print "    {\"%s\"," % tableName
+        if table.is_root:
+            is_root = "true"
+        else:
+            is_root = "false"
+        print "    {\"%s\", %s," % (tableName, is_root)
         print "     %s_columns, ARRAY_SIZE(%s_columns)," % (
             structName, structName)
         print "     sizeof(struct %s)}," % structName

ovsdb/ovsdb.c

@@ -127,6 +127,21 @@ is_valid_version(const char *s)
     return n != -1 && s[n] == '\0';
 }
 
+/* Returns the number of tables in 'schema''s root set. */
+static size_t
+root_set_size(const struct ovsdb_schema *schema)
+{
+    struct shash_node *node;
+    size_t n_root;
+
+    SHASH_FOR_EACH (node, &schema->tables) {
+        struct ovsdb_table_schema *table = node->data;
+
+        n_root += table->is_root;
+    }
+    return n_root;
+}
+
 struct ovsdb_error *
 ovsdb_schema_from_json(struct json *json, struct ovsdb_schema **schemap)
 {
@@ -205,6 +220,18 @@ ovsdb_schema_from_json(struct json *json, struct ovsdb_schema **schemap)
         }
     }
 
+    /* "isRoot" was not part of the original schema definition.  Before it was
+     * added, there was no support for garbage collection.  So, for backward
+     * compatibility, if the root set is empty then assume that every table is
+     * in the root set. */
+    if (root_set_size(schema) == 0) {
+        SHASH_FOR_EACH (node, &schema->tables) {
+            struct ovsdb_table_schema *table = node->data;
+
+            table->is_root = true;
+        }
+    }
+
     *schemap = schema;
     return 0;
 }
@@ -214,6 +241,7 @@ ovsdb_schema_to_json(const struct ovsdb_schema *schema)
 {
     struct json *json, *tables;
     struct shash_node *node;
+    bool default_is_root;
 
     json = json_object_create();
     json_object_put_string(json, "name", schema->name);
@@ -224,12 +252,18 @@ ovsdb_schema_to_json(const struct ovsdb_schema *schema)
         json_object_put_string(json, "cksum", schema->cksum);
     }
 
+    /* "isRoot" was not part of the original schema definition.  Before it was
+     * added, there was no support for garbage collection.  So, for backward
+     * compatibility, if every table is in the root set then do not output
+     * "isRoot" in table schemas. */
+    default_is_root = root_set_size(schema) == shash_count(&schema->tables);
+
     tables = json_object_create();
 
     SHASH_FOR_EACH (node, &schema->tables) {
         struct ovsdb_table_schema *table = node->data;
         json_object_put(tables, table->name,
-                        ovsdb_table_schema_to_json(table));
+                        ovsdb_table_schema_to_json(table, default_is_root));
     }
     json_object_put(json, "tables", tables);