sstables: introduce ms sstable format version

Introduce `ms` -- a new sstable format version which
is a hybrid of Cassandra's `me` and `da`.

It is based on `me`, but with the index components
(Summary.db and Index.db) replaced with the index
components of `da` (Partitions.db and Rows.db).

As of this patch, the version is never chosen
anywhere for writing sstables yet. It is only introduced.
We will add it to unit tests in a later commit,
and expose it to users in yet later commit.

api/api-doc/storage_service.json

@@ -3430,7 +3430,7 @@
             "version":{
                "type":"string",
                "enum":[
-                  "ka", "la", "mc", "md", "me"
+                  "ka", "la", "mc", "md", "me", "ms"
                ],
                "description":"SSTable version"
             },

docs/architecture/sstable/_common/sstable_what_is.rst

@@ -14,6 +14,9 @@ SSTable Version Support
    * - SSTable Version
      - ScyllaDB Enterprise Version
      - ScyllaDB Open Source Version
+   * - 3.x ('ms')
+     - 2025.4 and above
+     - None
    * - 3.x ('me')
      - 2022.2 and above
      - 5.1 and above

docs/architecture/sstable/sstable3/sstables-3-data-file-format.rst

@@ -35,6 +35,14 @@ Note that the file on-disk format applies to all "m*" SSTable format versions ("
 
 See :doc:`SSTables 3.0 Statistics File Format </architecture/sstable/sstable3/sstables-3-statistics>` for more details.
 
+Since Scylla 2025.4, there is a "ms" format in Scylla which is a hybrid of "me" and the "da" format introduced in Cassandra 5.0.
+Most components in "ms" are exactly the same as in "me", but the index components (Index.db and Summary.db) are replaced
+with the trie-based index format (components Partitions.db and Rows.db) used in "da".
+
+Partitions.db and Rows.db are not described on this page.
+See https://github.com/apache/cassandra/blob/70bcaec54d492658cb331b70869215ad64feb63d/src/java/org/apache/cassandra/io/sstable/format/bti/BtiFormat.md
+for a description.
+
 Overview
 ........
 

docs/dev/sstables-directory-structure.md

@@ -61,7 +61,7 @@ sub-directories as documented below.
 SSTables are comprised of multiple component files.
 The component file names are self-identifying and denote the component type, as well as per-sstable-format metadata.
 
-Here are the different component types and their naming convention:
+Here are the different component types:
 
 * Data (`Data.db`)  
   The SSTable data file, containing a part of the actual data stored in the database.
@@ -107,13 +107,29 @@ Here are the different component types and their naming convention:
 * Scylla (`Scylla.db`)  
   A file holding scylla-specific metadata about the SSTable, such as sharding information, extended features support, and sstabe-run identifier.
 
+
+* Partition Key Index (`Partitions.db`)  
+  Trie-based index of partition keys with pointers to their positions in the data file, or to a intra-partition index in Rows.db.
+
+
+* Clustering Key Index (`Rows.db`)  
+  Trie-based index of clustering keys within partitions. Used in conjunction with `Partitions.db`
+  as a replacement for `Index.db` and `Summary.db` in newest sstable formats.
+  (`da` in Cassandra, `ms` in Scylla).
+
+
+* Temporary partition key hashes (`TemporaryHashes.db`)  
+  A temporary file used for storing intermediate ingredients for the bloom filter.
+  This file appears only during write, and is deleted before the sstable is sealed.
+
 ### SSTable Format Version
 
 SSTable's on-disk format has changed over time.
-Three versions are currently supported by Scylla: `ka`, `la`, and `mc`.
+The versions currently supported by Scylla are: `ka`, `la`, `mc`, `md`, `me`, `ms`.
 Cassandra's convention is that the first letter determines
 the major format version, in ascending order, and the second letter -
 the minor version, starting from `a` onward.
+(`ms` is a Scylla-specific extension of `me`, so it breaks away from the `mc`-`me` series).
 
 The SSTable file names identify the SSTable format version.
 In addition, they provide the SSTable generation number and other metadata.
@@ -129,8 +145,10 @@ and it is version specific, as follows:
 
 where:
 * `<generation>` is the SSTable generation - a unique positive number identifying the SSTable.
-* `<big>` is an archaic attribute that identifies the SSTable sub-format.  
-  (Only `big` sub-format is supported by Scylla (and Cassandra) at this time.)
+* `<big>` is an attribute that identifies the SSTable sub-format.
+  (Only `big` sub-format is supported by Scylla at this time.
+  Cassandra 5.0 introduced `bti` (which stands for `BIG, trie-indexed`).
+  Version `ms` is a hybrid between `big` and `bti`).
 * `<component>` is the file's component type, as described above.
 
 ### Table Sub-directories

scylla-gdb.py

@@ -4546,7 +4546,8 @@ class scylla_sstables(gdb.Command):
             'la': new_format,
             'mc': new_format,
             'md': new_format,
-            'me': new_format
+            'me': new_format,
+            'ms': new_format,
         }
         format_to_str = ['big']
         schema = schema_ptr(sst['_schema'])

sstables/sstable_version.cc

@@ -37,20 +37,15 @@ const sstring sstable_version_constants::TEMPORARY_TOC_SUFFIX = "TOC.txt.tmp";
 
 sstable_version_constants::component_map_t sstable_version_constants::create_component_map() {
     return {
-        { component_type::Index, "Index.db"},
         { component_type::CompressionInfo, "CompressionInfo.db" },
         { component_type::Data, "Data.db" },
         { component_type::TOC, TOC_SUFFIX },
-        { component_type::Summary, "Summary.db" },
         { component_type::CRC, "CRC.db" },
         { component_type::Filter, "Filter.db" },
         { component_type::Statistics, "Statistics.db" },
         { component_type::Scylla, "Scylla.db" },
         { component_type::TemporaryTOC, TEMPORARY_TOC_SUFFIX },
-        { component_type::TemporaryStatistics, "Statistics.db.tmp" },
-        { component_type::Rows, "Rows.db" },
-        { component_type::Partitions, "Partitions.db" },
-        { component_type::TemporaryHashes, "TemporaryHashes.db.tmp" },
+        { component_type::TemporaryStatistics, "Statistics.db.tmp" }
     };
 }
 
@@ -64,6 +59,8 @@ sstable_version_constants::get_component_map(sstable_version_types version) {
         case sstable_version_types::md:
         case sstable_version_types::me:
             return sstable_version_constants_m::_component_map;
+        case sstable_version_types::ms:
+            return sstable_version_constants_ms::_component_map;
     }
     // Should never reach this.
     // Compiler should complain if the switch above does no cover all sstable_version_types values.
@@ -72,6 +69,8 @@ sstable_version_constants::get_component_map(sstable_version_types version) {
 
 const sstable_version_constants::component_map_t sstable_version_constants_k_l::create_component_map() {
     auto result = sstable_version_constants::create_component_map();
+    result.emplace(component_type::Index, "Index.db");
+    result.emplace(component_type::Summary, "Summary.db");
     result.emplace(component_type::Digest, "Digest.sha1");
     return result;
 }
@@ -81,6 +80,8 @@ const sstable_version_constants::component_map_t sstable_version_constants_k_l::
 
 const sstable_version_constants::component_map_t sstable_version_constants_m::create_component_map() {
     auto result = sstable_version_constants::create_component_map();
+    result.emplace(component_type::Index, "Index.db");
+    result.emplace(component_type::Summary, "Summary.db");
     result.emplace(component_type::Digest, "Digest.crc32");
     return result;
 }
@@ -88,4 +89,17 @@ const sstable_version_constants::component_map_t sstable_version_constants_m::cr
 const sstable_version_constants::component_map_t sstable_version_constants_m::_component_map =
         sstable_version_constants_m::create_component_map();
 
+const sstable_version_constants::component_map_t sstable_version_constants_ms::create_component_map() {
+    auto result = sstable_version_constants_m::create_component_map();
+    // Note: for `ms`, we inherit all components from `me`.
+    // This means that we allow `ms` to have Index.db and Summary.db components.
+    result.emplace(component_type::Rows, "Rows.db");
+    result.emplace(component_type::Partitions, "Partitions.db");
+    result.emplace(component_type::TemporaryHashes, "TemporaryHashes.db.tmp");
+    return result;
 }
+
+const sstable_version_constants::component_map_t sstable_version_constants_ms::_component_map =
+        sstable_version_constants_ms::create_component_map();
+
+}

sstables/sstable_version_m.hh

@@ -14,10 +14,17 @@
 namespace sstables {
 
 class sstable_version_constants_m final : public sstable_version_constants {
-    static const sstable_version_constants::component_map_t create_component_map();
 public:
+    static const sstable_version_constants::component_map_t create_component_map();
     sstable_version_constants_m() = delete;
     static const sstable_version_constants::component_map_t _component_map;
 };
 
+class sstable_version_constants_ms final : public sstable_version_constants {
+public:
+    static const sstable_version_constants::component_map_t create_component_map();
+    sstable_version_constants_ms() = delete;
+    static const sstable_version_constants::component_map_t _component_map;
+};
+
 }

sstables/sstables.cc

@@ -197,6 +197,7 @@ const std::unordered_map<sstable_version_types, sstring, enum_hash<sstable_versi
     { sstable_version_types::mc , "mc" },
     { sstable_version_types::md , "md" },
     { sstable_version_types::me , "me" },
+    { sstable_version_types::ms , "ms" },
 };
 
 const std::unordered_map<sstable_format_types, sstring, enum_hash<sstable_format_types>> format_string = {
@@ -2447,6 +2448,7 @@ sstring sstable::component_basename(const sstring& ks, const sstring& cf, versio
     case sstable::version_types::mc:
     case sstable::version_types::md:
     case sstable::version_types::me:
+    case sstable::version_types::ms:
         return v + "-" + g + "-" + f + "-" + component;
     }
     on_internal_error(sstlog, seastar::format("invalid version {} for sstable: table={}.{}, generation={}, format={}, component={}",
@@ -2586,7 +2588,7 @@ static std::tuple<entry_descriptor, sstring, sstring> make_entry_descriptor(cons
     //   la-42-big-Data.db
     //   ka-42-big-Data.db
     //   me-3g8w_00qf_4pbog2i7h2c7am0uoe-big-Data.db
-    static boost::regex la_mx("(la|m[cde])-([^-]+)-(\\w+)-(.*)");
+    static boost::regex la_mx("(la|m[cdes])-([^-]+)-(\\w+)-(.*)");
     static boost::regex ka("(\\w+)-(\\w+)-ka-(\\d+)-(.*)");
 
     // Use non-greedy match so that a snapshot tag that ressembles a name-<uuid> wouldn't match
@@ -2670,6 +2672,10 @@ sstable_format_types format_from_string(std::string_view s) {
     }
 }
 
+bool has_summary_and_index(sstable_version_types v) {
+    return v != sstable_version_types::ms;
+}
+
 component_type sstable::component_from_sstring(version_types v, const sstring &s) {
     try {
         return reverse_map(s, sstable_version_constants::get_component_map(v));

sstables/types.hh

@@ -286,6 +286,7 @@ struct compaction_metadata : public metadata_base<compaction_metadata> {
         case sstable_version_types::mc:
         case sstable_version_types::md:
         case sstable_version_types::me:
+        case sstable_version_types::ms:
             return f(
                 cardinality
             );
@@ -331,6 +332,7 @@ struct stats_metadata : public metadata_base<stats_metadata> {
     template <typename Describer>
     auto describe_type(sstable_version_types v, Describer f) {
         switch (v) {
+        case sstable_version_types::ms:
         case sstable_version_types::me:
             return f(
                 estimated_partition_size,
@@ -429,6 +431,7 @@ struct serialization_header : public metadata_base<serialization_header> {
         case sstable_version_types::mc:
         case sstable_version_types::md:
         case sstable_version_types::me:
+        case sstable_version_types::ms:
             return f(
                 min_timestamp_base,
                 min_local_deletion_time_base,

sstables/version.hh

@@ -14,7 +14,7 @@
 
 namespace sstables {
 
-enum class sstable_version_types { ka, la, mc, md, me };
+enum class sstable_version_types { ka, la, mc, md, me, ms };
 enum class sstable_format_types { big };
 
 constexpr std::array<sstable_version_types, 5> all_sstable_versions = {
@@ -23,12 +23,18 @@ constexpr std::array<sstable_version_types, 5> all_sstable_versions = {
     sstable_version_types::mc,
     sstable_version_types::md,
     sstable_version_types::me,
+    // FIXME: Uncomment after tests are prepared for the new
+    // version. This will happen in the same series.
+    // sstable_version_types::ms,
 };
 
 constexpr std::array<sstable_version_types, 3> writable_sstable_versions = {
     sstable_version_types::mc,
     sstable_version_types::md,
     sstable_version_types::me,
+    // FIXME: Uncomment after tests are prepared for the new
+    // version. This will happen in the same series.
+    // sstable_version_types::ms,
 };
 
 constexpr sstable_version_types oldest_writable_sstable_format = sstable_version_types::mc;
@@ -40,6 +46,8 @@ inline auto get_highest_sstable_version() {
 sstable_version_types version_from_string(std::string_view s);
 sstable_format_types format_from_string(std::string_view s);
 
+bool has_summary_and_index(sstable_version_types v);
+
 extern const std::unordered_map<sstable_version_types, seastar::sstring, seastar::enum_hash<sstable_version_types>> version_string;
 extern const std::unordered_map<sstable_format_types, seastar::sstring, seastar::enum_hash<sstable_format_types>> format_string;
 

tools/scylla-sstable.cc

@@ -2372,7 +2372,7 @@ directly expressed in CQL. Editing schema options (the part after WITH) is safe.
 
 The sstable version can be selected manually with the --sstable-version option,
 by default the latest supported version is used. Valid options are sstable
-versions which are supported for writing: mc, md and me.
+versions which are supported for writing: mc, md, me, ms.
 
 Mapping of input sstables to output sstables is printed to stdout.