e4865bd4d1
Currently, scans are splitting partition ranges around tokens. This will have to change with tablets, where we should split at tablet boundaries. This patch introduces token_range_splitter which abstracts this task. It is provided by effective_replication_map implementation.
42 lines
1.2 KiB
C++
42 lines
1.2 KiB
C++
/*
|
|
* Copyright (C) 2023-present ScyllaDB
|
|
*/
|
|
|
|
/*
|
|
* SPDX-License-Identifier: AGPL-3.0-or-later
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
#include "dht/i_partitioner.hh"
|
|
#include "dht/token.hh"
|
|
|
|
#include <optional>
|
|
|
|
namespace locator {
|
|
|
|
/// Generates split points which divide the ring into ranges which share the same replica set.
|
|
///
|
|
/// Initially the ring space the splitter works with is set to the whole ring.
|
|
/// The space can be changed using reset().
|
|
class token_range_splitter {
|
|
public:
|
|
virtual ~token_range_splitter() = default;
|
|
|
|
/// Resets the splitter to work with the ring range [pos, +inf).
|
|
virtual void reset(dht::ring_position_view pos) = 0;
|
|
|
|
/// Each token t returned by next_token() means that keys in the range:
|
|
///
|
|
/// [prev_pos, dht::ring_position_view::ending_at(t))
|
|
///
|
|
/// share the same replica set.
|
|
///
|
|
/// If this is the first call to next_token() after construction or reset() then prev_pos is the
|
|
/// beginning of the ring space. Otherwise, it is dht::ring_position_view::ending_at(prev_t)
|
|
/// where prev_t is the token returned by the previous call to next_token().
|
|
/// If std::nullopt is returned it means that the ring space was exhausted.
|
|
virtual std::optional<dht::token> next_token() = 0;
|
|
};
|
|
|
|
} |