feat(clp-s): Record log-order at compression time.

components/core/src/clp_s/ArchiveReader.cpp

@@ -27,6 +27,8 @@ void ArchiveReader::open(string_view archives_dir, string_view archive_id) {
     m_schema_tree = ReaderUtils::read_schema_tree(archive_path_str);
     m_schema_map = ReaderUtils::read_schemas(archive_path_str);
 
+    m_log_event_idx_column_id = m_schema_tree->get_internal_field_id(constants::cLogEventIdxName);
+
     m_table_metadata_file_reader.open(archive_path_str + constants::cArchiveTableMetadataFile);
     m_stream_reader.open_packed_streams(archive_path_str + constants::cArchiveTablesFile);
 }
@@ -310,6 +312,12 @@ void ArchiveReader::initialize_schema_reader(
         }
         BaseColumnReader* column_reader = append_reader_column(reader, column_id);
 
+        if (column_id == m_log_event_idx_column_id
+            && nullptr != dynamic_cast<Int64ColumnReader*>(column_reader))
+        {
+            reader.mark_column_as_log_event_idx(static_cast<Int64ColumnReader*>(column_reader));
+        }
+
         if (should_extract_timestamp && column_reader && timestamp_column_ids.count(column_id) > 0)
         {
             reader.mark_column_as_timestamp(column_reader);
@@ -346,6 +354,7 @@ void ArchiveReader::close() {
     m_cur_stream_id = 0;
     m_stream_buffer.reset();
     m_stream_buffer_size = 0ULL;
+    m_log_event_idx_column_id = -1;
 }
 
 std::shared_ptr<char[]> ArchiveReader::read_stream(size_t stream_id, bool reuse_buffer) {

components/core/src/clp_s/ArchiveReader.hpp

@@ -214,6 +214,7 @@ class ArchiveReader {
     std::shared_ptr<char[]> m_stream_buffer{};
     size_t m_stream_buffer_size{0ULL};
     size_t m_cur_stream_id{0ULL};
+    int32_t m_log_event_idx_column_id{-1};
 };
 }  // namespace clp_s
 

components/core/src/clp_s/ArchiveWriter.cpp

@@ -68,6 +68,7 @@ void ArchiveWriter::close() {
     m_encoded_message_size = 0UL;
     m_uncompressed_size = 0UL;
     m_compressed_size = 0UL;
+    m_next_log_event_id = 0;
 }
 
 void ArchiveWriter::append_message(
@@ -86,6 +87,7 @@ void ArchiveWriter::append_message(
     }
 
     m_encoded_message_size += schema_writer->append_message(message);
+    ++m_next_log_event_id;
 }
 
 size_t ArchiveWriter::get_data_size() {

components/core/src/clp_s/ArchiveWriter.hpp

@@ -1,6 +1,7 @@
 #ifndef CLP_S_ARCHIVEWRITER_HPP
 #define CLP_S_ARCHIVEWRITER_HPP
 
+#include <string_view>
 #include <utility>
 
 #include <boost/filesystem.hpp>
@@ -93,10 +94,15 @@ class ArchiveWriter {
      * @param key
      * @return the node id
      */
-    int32_t add_node(int parent_node_id, NodeType type, std::string const& key) {
+    int32_t add_node(int parent_node_id, NodeType type, std::string_view const key) {
         return m_schema_tree.add_node(parent_node_id, type, key);
     }
 
+    /**
+     * @return the Id the next log event should receive when appended to the archive.
+     */
+    int64_t get_next_log_event_id() const { return m_next_log_event_id; }
+
     /**
      * Return a schema's Id and add the schema to the
      * schema map if it does not already exist.
@@ -174,6 +180,7 @@ class ArchiveWriter {
     size_t m_encoded_message_size{};
     size_t m_uncompressed_size{};
     size_t m_compressed_size{};
+    int64_t m_next_log_event_id{};
 
     std::string m_id;
 

components/core/src/clp_s/CommandLineArguments.cpp

@@ -296,13 +296,13 @@ CommandLineArguments::parse_arguments(int argc, char const** argv) {
             decompression_options.add_options()(
                     "ordered",
                     po::bool_switch(&m_ordered_decompression),
-                    "Enable decompression in ascending timestamp order for this archive"
+                    "Enable decompression in log order for this archive"
             )(
                     "ordered-chunk-size",
                     po::value<size_t>(&m_ordered_chunk_size)
                             ->default_value(m_ordered_chunk_size),
                     "Number of records to include in each output file when decompressing records "
-                    "in ascending timestamp order"
+                    "in log order"
             );
             // clang-format on
             extraction_options.add(decompression_options);

components/core/src/clp_s/JsonConstructor.cpp

@@ -68,15 +68,15 @@ void JsonConstructor::construct_in_order() {
     auto tables = m_archive_reader->read_all_tables();
     using ReaderPointer = std::shared_ptr<SchemaReader>;
     auto cmp = [](ReaderPointer& left, ReaderPointer& right) {
-        return left->get_next_timestamp() > right->get_next_timestamp();
+        return left->get_next_log_event_idx() > right->get_next_log_event_idx();
     };
     std::priority_queue record_queue(tables.begin(), tables.end(), cmp);
     // Clear tables vector so that memory gets deallocated after we have marshalled all records for
     // a given table
     tables.clear();
 
-    epochtime_t first_timestamp{0};
-    epochtime_t last_timestamp{0};
+    int64_t first_idx{0};
+    int64_t last_idx{0};
     size_t num_records_marshalled{0};
     auto src_path = std::filesystem::path(m_option.output_dir) / m_option.archive_id;
     FileWriter writer;
@@ -98,8 +98,8 @@ void JsonConstructor::construct_in_order() {
     std::vector<bsoncxx::document::value> results;
     auto finalize_chunk = [&](bool open_new_writer) {
         writer.close();
-        std::string new_file_name = src_path.string() + "_" + std::to_string(first_timestamp) + "_"
-                                    + std::to_string(last_timestamp) + ".jsonl";
+        std::string new_file_name = src_path.string() + "_" + std::to_string(first_idx) + "_"
+                                    + std::to_string(last_idx) + ".jsonl";
         auto new_file_path = std::filesystem::path(new_file_name);
         std::error_code ec;
         std::filesystem::rename(src_path, new_file_path, ec);
@@ -119,11 +119,11 @@ void JsonConstructor::construct_in_order() {
                     ),
                     bsoncxx::builder::basic::kvp(
                             constants::results_cache::decompression::cBeginMsgIx,
-                            static_cast<int64_t>(first_timestamp)
+                            first_idx
                     ),
                     bsoncxx::builder::basic::kvp(
                             constants::results_cache::decompression::cEndMsgIx,
-                            static_cast<int64_t>(last_timestamp)
+                            last_idx
                     ),
                     bsoncxx::builder::basic::kvp(
                             constants::results_cache::decompression::cIsLastIrChunk,
@@ -140,9 +140,9 @@ void JsonConstructor::construct_in_order() {
     while (false == record_queue.empty()) {
         ReaderPointer next = record_queue.top();
         record_queue.pop();
-        last_timestamp = next->get_next_timestamp();
+        last_idx = next->get_next_log_event_idx();
         if (0 == num_records_marshalled) {
-            first_timestamp = last_timestamp;
+            first_idx = last_idx;
         }
         next->get_next_message(buffer);
         if (false == next->done()) {

components/core/src/clp_s/JsonConstructor.hpp

@@ -66,7 +66,7 @@ class JsonConstructor {
 private:
     /**
      * Reads all of the tables from m_archive_reader and writes all of the records
-     * they contain to writer in timestamp order.
+     * they contain to writer in log order.
      */
     void construct_in_order();
 

components/core/src/clp_s/JsonParser.cpp

@@ -462,11 +462,19 @@ bool JsonParser::parse() {
                 return false;
             }
 
+            // Add internal log_event_idx field to record
+            auto log_event_idx = add_internal_field(constants::cLogEventIdxName, NodeType::Integer);
+            m_current_parsed_message.add_value(
+                    log_event_idx,
+                    m_archive_writer->get_next_log_event_id()
+            );
+            m_current_schema.insert_ordered(log_event_idx);
+
             // Some errors from simdjson are latent until trying to access invalid JSON fields.
             // Instead of checking for an error every time we access a JSON field in parse_line we
             // just catch simdjson_error here instead.
             try {
-                parse_line(ref.value(), -1, "");
+                parse_line(ref.value(), constants::cRootNodeId, constants::cRootNodeName);
             } catch (simdjson::simdjson_error& error) {
                 SPDLOG_ERROR(
                         "Encountered error - {} - while trying to parse {} after parsing {} bytes",
@@ -521,6 +529,15 @@ bool JsonParser::parse() {
     return true;
 }
 
+int32_t JsonParser::add_internal_field(std::string_view const field_name, NodeType type) {
+    auto internal_subtree_id = m_archive_writer->add_node(
+            constants::cRootNodeId,
+            NodeType::Internal,
+            constants::cInternalSubtreeName
+    );
+    return m_archive_writer->add_node(internal_subtree_id, type, field_name);
+}
+
 void JsonParser::store() {
     m_archive_writer->close();
 }

components/core/src/clp_s/JsonParser.hpp

@@ -3,6 +3,7 @@
 
 #include <map>
 #include <string>
+#include <string_view>
 #include <variant>
 #include <vector>
 
@@ -94,6 +95,14 @@ class JsonParser {
      */
     void split_archive();
 
+    /**
+     * Add an internal field to the MPT and get its Id.
+     *
+     * Note: this method should be called before parsing a record so that internal fields come first
+     * in each table. This isn't strictly necessary, but it is a nice convention.
+     */
+    int32_t add_internal_field(std::string_view const field_name, NodeType type);
+
     int m_num_messages;
     std::vector<std::string> m_file_paths;
 

components/core/src/clp_s/JsonSerializer.hpp

@@ -2,6 +2,7 @@
 #define CLP_S_JSONSERIALIZER_HPP
 
 #include <string>
+#include <string_view>
 #include <vector>
 
 #include "ColumnReader.hpp"
@@ -66,7 +67,7 @@ class JsonSerializer {
         return false;
     }
 
-    void add_special_key(std::string const& key) { m_special_keys.push_back(key); }
+    void add_special_key(std::string_view const key) { m_special_keys.emplace_back(key); }
 
     void begin_object() {
         append_key();
@@ -110,13 +111,13 @@ class JsonSerializer {
 
     void append_key() { append_key(m_special_keys[m_special_keys_index++]); }
 
-    void append_key(std::string const& key) {
+    void append_key(std::string_view const key) {
         m_json_string += "\"";
         m_json_string += key;
         m_json_string += "\":";
     }
 
-    void append_value(std::string const& value) {
+    void append_value(std::string_view const value) {
         m_json_string += value;
         m_json_string += ",";
     }

components/core/src/clp_s/ReaderUtils.cpp

@@ -18,10 +18,10 @@ std::shared_ptr<SchemaTree> ReaderUtils::read_schema_tree(std::string const& arc
         throw OperationFailed(error_code, __FILENAME__, __LINE__);
     }
 
+    std::string key;
     for (size_t i = 0; i < num_nodes; i++) {
         int32_t parent_id;
         size_t key_length;
-        std::string key;
         uint8_t node_type;
 
         error_code = schema_tree_decompressor.try_read_numeric_value(parent_id);

components/core/src/clp_s/SchemaReader.cpp

@@ -37,6 +37,13 @@ void SchemaReader::mark_column_as_timestamp(BaseColumnReader* column_reader) {
     }
 }
 
+int64_t SchemaReader::get_next_log_event_idx() const {
+    if (nullptr != m_log_event_idx_column) {
+        return std::get<int64_t>(m_log_event_idx_column->extract_value(m_cur_message));
+    }
+    return 0;
+}
+
 void SchemaReader::load(
         std::shared_ptr<char[]> stream_buffer,
         size_t offset,
@@ -86,7 +93,7 @@ void SchemaReader::generate_json_string() {
             }
             case JsonSerializer::Op::AddIntField: {
                 column = m_reordered_columns[column_id_index++];
-                auto const& name = m_global_schema_tree->get_node(column->get_id()).get_key_name();
+                auto name = m_global_schema_tree->get_node(column->get_id()).get_key_name();
                 m_json_serializer.append_key(name);
                 m_json_serializer.append_value(
                         std::to_string(std::get<int64_t>(column->extract_value(m_cur_message)))
@@ -102,7 +109,7 @@ void SchemaReader::generate_json_string() {
             }
             case JsonSerializer::Op::AddFloatField: {
                 column = m_reordered_columns[column_id_index++];
-                auto const& name = m_global_schema_tree->get_node(column->get_id()).get_key_name();
+                auto name = m_global_schema_tree->get_node(column->get_id()).get_key_name();
                 m_json_serializer.append_key(name);
                 m_json_serializer.append_value(
                         std::to_string(std::get<double>(column->extract_value(m_cur_message)))
@@ -118,7 +125,7 @@ void SchemaReader::generate_json_string() {
             }
             case JsonSerializer::Op::AddBoolField: {
                 column = m_reordered_columns[column_id_index++];
-                auto const& name = m_global_schema_tree->get_node(column->get_id()).get_key_name();
+                auto name = m_global_schema_tree->get_node(column->get_id()).get_key_name();
                 m_json_serializer.append_key(name);
                 m_json_serializer.append_value(
                         std::get<uint8_t>(column->extract_value(m_cur_message)) != 0 ? "true"
@@ -136,7 +143,7 @@ void SchemaReader::generate_json_string() {
             }
             case JsonSerializer::Op::AddStringField: {
                 column = m_reordered_columns[column_id_index++];
-                auto const& name = m_global_schema_tree->get_node(column->get_id()).get_key_name();
+                auto name = m_global_schema_tree->get_node(column->get_id()).get_key_name();
                 m_json_serializer.append_key(name);
                 m_json_serializer.append_value_from_column_with_quotes(column, m_cur_message);
                 break;
@@ -215,9 +222,10 @@ bool SchemaReader::get_next_message(std::string& message, FilterClass* filter) {
     return false;
 }
 
-bool SchemaReader::get_next_message_with_timestamp(
+bool SchemaReader::get_next_message_with_metadata(
         std::string& message,
         epochtime_t& timestamp,
+        int64_t& log_event_idx,
         FilterClass* filter
 ) {
     // TODO: If we already get max_num_results messages, we can skip messages
@@ -241,6 +249,7 @@ bool SchemaReader::get_next_message_with_timestamp(
         }
 
         timestamp = m_get_timestamp();
+        log_event_idx = get_next_log_event_idx();
 
         m_cur_message++;
         return true;
@@ -561,7 +570,7 @@ void SchemaReader::initialize_serializer() {
     // TODO: this code will have to change once we allow mixing log lines parsed by different
     // parsers.
     if (false == m_local_schema_tree.get_nodes().empty()) {
-        generate_json_template(m_local_schema_tree.get_root_node_id());
+        generate_json_template(m_local_schema_tree.get_object_subtree_node_id());
     }
 }
 
@@ -572,7 +581,7 @@ void SchemaReader::generate_json_template(int32_t id) {
     for (int32_t child_id : children_ids) {
         int32_t child_global_id = m_local_id_to_global_id[child_id];
         auto const& child_node = m_local_schema_tree.get_node(child_id);
-        std::string const& key = child_node.get_key_name();
+        auto key = child_node.get_key_name();
         switch (child_node.get_type()) {
             case NodeType::Object: {
                 m_json_serializer.add_op(JsonSerializer::Op::BeginObject);