scylladb/sstables/compaction.cc

/*
 * Copyright (C) 2015 ScyllaDB
 */

/*
 * This file is part of Scylla.
 *
 * Scylla is free software: you can redistribute it and/or modify
 * it under the terms of the GNU Affero General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * Scylla is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
 */

/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you under the Apache License, Version 2.0 (the
 * "License"); you may not use this file except in compliance
 * with the License.  You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <vector>
#include <map>
#include <functional>
#include <utility>
#include <assert.h>
#include <algorithm>

#include <boost/range/algorithm.hpp>
#include <boost/range/adaptors.hpp>
#include <boost/range/join.hpp>
#include <boost/algorithm/cxx11/any_of.hpp>

#include "core/future-util.hh"
#include "core/pipe.hh"

#include "sstables.hh"
#include "sstables/progress_monitor.hh"
#include "compaction.hh"
#include "compaction_manager.hh"
#include "database.hh"
#include "mutation_reader.hh"
#include "schema.hh"
#include "db/system_keyspace.hh"
#include "service/storage_service.hh"
#include "service/priority_manager.hh"
#include "db_clock.hh"
#include "mutation_compactor.hh"
#include "leveled_manifest.hh"

namespace sstables {

logging::logger clogger("compaction");

static api::timestamp_type get_max_purgeable_timestamp(const column_family& cf, sstable_set::incremental_selector& selector,
        const std::unordered_set<shared_sstable>& compacting_set, const dht::decorated_key& dk) {
    auto timestamp = api::max_timestamp;
    stdx::optional<utils::hashed_key> hk;
    for (auto&& sst : boost::range::join(selector.select(dk.token()).sstables, cf.compacted_undeleted_sstables())) {
        if (compacting_set.count(sst)) {
            continue;
        }
        if (!hk) {
            hk = sstables::sstable::make_hashed_key(*cf.schema(), dk.key());
        }
        if (sst->filter_has_key(*hk)) {
            timestamp = std::min(timestamp, sst->get_stats_metadata().min_timestamp);
        }
    }
    return timestamp;
}

static bool belongs_to_current_node(const dht::token& t, const dht::token_range_vector& sorted_owned_ranges) {
    auto low = std::lower_bound(sorted_owned_ranges.begin(), sorted_owned_ranges.end(), t,
            [] (const range<dht::token>& a, const dht::token& b) {
        // check that range a is before token b.
        return a.after(b, dht::token_comparator());
    });

    if (low != sorted_owned_ranges.end()) {
        const dht::token_range& r = *low;
        return r.contains(t, dht::token_comparator());
    }

    return false;
}

static void delete_sstables_for_interrupted_compaction(std::vector<shared_sstable>& new_sstables, sstring& ks, sstring& cf) {
    // Delete either partially or fully written sstables of a compaction that
    // was either stopped abruptly (e.g. out of disk space) or deliberately
    // (e.g. nodetool stop COMPACTION).
    for (auto& sst : new_sstables) {
        clogger.debug("Deleting sstable {} of interrupted compaction for {}.{}", sst->get_filename(), ks, cf);
        sst->mark_for_deletion();
    }
}

static std::vector<shared_sstable> get_uncompacting_sstables(column_family& cf, std::vector<shared_sstable> sstables) {
    auto all_sstables = boost::copy_range<std::vector<shared_sstable>>(*cf.get_sstables_including_compacted_undeleted());
    boost::sort(all_sstables, [] (const shared_sstable& x, const shared_sstable& y) {
        return x->generation() < y->generation();
    });
    std::sort(sstables.begin(), sstables.end(), [] (const shared_sstable& x, const shared_sstable& y) {
        return x->generation() < y->generation();
    });
    std::vector<shared_sstable> not_compacted_sstables;
    boost::set_difference(all_sstables, sstables,
        std::back_inserter(not_compacted_sstables), [] (const shared_sstable& x, const shared_sstable& y) {
            return x->generation() < y->generation();
        });
    return not_compacted_sstables;
}

class compaction;

class compacting_sstable_writer {
    compaction& _c;
    sstable_writer* _writer = nullptr;
public:
    explicit compacting_sstable_writer(compaction& c) : _c(c) {}

    void consume_new_partition(const dht::decorated_key& dk);

    void consume(tombstone t) { _writer->consume(t); }
    stop_iteration consume(static_row&& sr, tombstone, bool) { return _writer->consume(std::move(sr)); }
    stop_iteration consume(clustering_row&& cr, row_tombstone, bool) { return _writer->consume(std::move(cr)); }
    stop_iteration consume(range_tombstone&& rt) { return _writer->consume(std::move(rt)); }

    stop_iteration consume_end_of_partition();
    void consume_end_of_stream();
};

struct compaction_read_monitor_generator final : public read_monitor_generator {
    class compaction_read_monitor final : public  sstables::read_monitor, public backlog_read_progress_manager {
        sstables::shared_sstable _sst;
        compaction_manager& _compaction_manager;
        column_family& _cf;
        const sstables::reader_position_tracker* _tracker = nullptr;
        uint64_t _last_position_seen = 0;
    public:
        virtual void on_read_started(const sstables::reader_position_tracker& tracker) override {
            _tracker = &tracker;
            _cf.get_compaction_strategy().get_backlog_tracker().register_compacting_sstable(_sst, *this);
        }

        virtual void on_read_completed() override {
            if (_tracker) {
                _last_position_seen = _tracker->position;
                _tracker = nullptr;
            }
        }

        virtual uint64_t compacted() const override {
            if (_tracker) {
                return _tracker->position;
            }
            return _last_position_seen;
        }

        void remove_sstable(bool is_tracking) {
            if (is_tracking) {
                _cf.get_compaction_strategy().get_backlog_tracker().remove_sstable(_sst);
            }
            _sst = {};
        }

        compaction_read_monitor(sstables::shared_sstable sst, compaction_manager& cm, column_family &cf)
            : _sst(std::move(sst)), _compaction_manager(cm), _cf(cf) { }

        ~compaction_read_monitor() {
            // We failed to finish handling this SSTable, so we have to update the backlog_tracker
            // about it.
            if (_sst) {
                _cf.get_compaction_strategy().get_backlog_tracker().revert_charges(_sst);
            }
        }
    };

    virtual sstables::read_monitor& operator()(sstables::shared_sstable sst) override {
        _generated_monitors.emplace_back(std::move(sst), _compaction_manager, _cf);
        return _generated_monitors.back();
    }
    compaction_read_monitor_generator(compaction_manager& cm, column_family& cf)
        : _compaction_manager(cm)
        , _cf(cf) {}

    void remove_sstables(bool is_tracking) {
        for (auto& rm : _generated_monitors) {
            rm.remove_sstable(is_tracking);
        }

    }
private:
     compaction_manager& _compaction_manager;
     column_family& _cf;
     std::deque<compaction_read_monitor> _generated_monitors;
};

class compaction_write_monitor final : public sstables::write_monitor, public backlog_write_progress_manager {
    sstables::shared_sstable _sst;
    column_family& _cf;
    const sstables::writer_offset_tracker* _tracker = nullptr;
    uint64_t _progress_seen = 0;
public:
    compaction_write_monitor(sstables::shared_sstable sst, column_family& cf)
        : _sst(sst)
        , _cf(cf) {}

    ~compaction_write_monitor() {
        if (_sst) {
            _cf.get_compaction_strategy().get_backlog_tracker().revert_charges(_sst);
        }
    }

    virtual void on_write_started(const sstables::writer_offset_tracker& tracker) override {
        _tracker = &tracker;
        _cf.get_compaction_strategy().get_backlog_tracker().register_partially_written_sstable(_sst, *this);
    }

    virtual void on_data_write_completed() override {
        if (_tracker) {
            _progress_seen = _tracker->offset;
            _tracker = nullptr;
        }
    }

    virtual uint64_t written() const {
        if (_tracker) {
            return _tracker->offset;
        }
        return _progress_seen;
    }

    void add_sstable() {
        _cf.get_compaction_strategy().get_backlog_tracker().add_sstable(_sst);
        _sst = {};
    }

    virtual void on_write_completed() override { }
    virtual void on_flush_completed() override { }
};

// Resharding doesn't really belong into any strategy, because it is not worried about laying out
// SSTables according to any strategy-specific criteria.  So we will just make it proportional to
// the amount of data we still have to reshard.
//
// Although at first it may seem like we could improve this by tracking the ongoing reshard as well
// and reducing the backlog as we compact, that is not really true. Resharding is not really
// expected to get rid of data and it is usually just splitting data among shards. Whichever backlog
// we get rid of by tracking the compaction will come back as a big spike as we add this SSTable
// back to their rightful shard owners.
//
// So because the data is supposed to be constant, we will just add the total amount of data as the
// backlog.
class resharding_backlog_tracker final : public compaction_backlog_tracker::impl {
    uint64_t _total_bytes = 0;
public:
    virtual double backlog(const compaction_backlog_tracker::ongoing_writes& ow, const compaction_backlog_tracker::ongoing_compactions& oc) const override {
        return _total_bytes;
    }

    virtual void add_sstable(sstables::shared_sstable sst)  override {
        _total_bytes += sst->data_size();
    }

    virtual void remove_sstable(sstables::shared_sstable sst)  override {
        _total_bytes -= sst->data_size();
    }
};

class compaction {
protected:
    column_family& _cf;
    std::vector<shared_sstable> _sstables;
    uint64_t _max_sstable_size;
    uint32_t _sstable_level;
    lw_shared_ptr<compaction_info> _info = make_lw_shared<compaction_info>();
    uint64_t _estimated_partitions = 0;
    std::vector<unsigned long> _ancestors;
    db::replay_position _rp;
    seastar::thread_scheduling_group* _tsg;
protected:
    compaction(column_family& cf, std::vector<shared_sstable> sstables, uint64_t max_sstable_size, uint32_t sstable_level, seastar::thread_scheduling_group* tsg)
        : _cf(cf)
        , _sstables(std::move(sstables))
        , _max_sstable_size(max_sstable_size)
        , _sstable_level(sstable_level)
        , _tsg(tsg)
    {
        _cf.get_compaction_manager().register_compaction(_info);
    }

    uint64_t partitions_per_sstable() const {
        uint64_t estimated_sstables = std::max(1UL, uint64_t(ceil(double(_info->start_size) / _max_sstable_size)));
        return ceil(double(_estimated_partitions) / estimated_sstables);
    }

    void setup_new_sstable(shared_sstable& sst) {
        _info->new_sstables.push_back(sst);
        sst->get_metadata_collector().set_replay_position(_rp);
        sst->get_metadata_collector().sstable_level(_sstable_level);
        for (auto ancestor : _ancestors) {
            sst->add_ancestor(ancestor);
        }
    }

    void finish_new_sstable(stdx::optional<sstable_writer>& writer, shared_sstable& sst) {
        writer->consume_end_of_stream();
        writer = stdx::nullopt;
        sst->open_data().get0();
        _info->end_size += sst->bytes_on_disk();
    }
public:
    compaction& operator=(const compaction&) = delete;
    compaction(const compaction&) = delete;

    virtual ~compaction() {
        if (_info) {
            _cf.get_compaction_manager().deregister_compaction(_info);
        }
    }

    seastar::thread_attributes thread_attributes() {
        seastar::thread_attributes attr;
        attr.scheduling_group = _tsg;
        return attr;
    }
private:
    // Default range sstable reader that will only return mutation that belongs to current shard.
    virtual flat_mutation_reader make_sstable_reader(lw_shared_ptr<sstables::sstable_set> ssts) const = 0;

    flat_mutation_reader setup() {
        auto ssts = make_lw_shared<sstables::sstable_set>(_cf.get_compaction_strategy().make_sstable_set(_cf.schema()));
        auto schema = _cf.schema();
        sstring formatted_msg = "[";
        auto fully_expired = get_fully_expired_sstables(_cf, _sstables, gc_clock::now() - schema->gc_grace_seconds());

        for (auto& sst : _sstables) {
            // Compacted sstable keeps track of its ancestors.
            _ancestors.push_back(sst->generation());
            _info->start_size += sst->bytes_on_disk();
            _info->total_partitions += sst->get_estimated_key_count();
            formatted_msg += sprint("%s:level=%d, ", sst->get_filename(), sst->get_sstable_level());

            // Do not actually compact a sstable that is fully expired and can be safely
            // dropped without ressurrecting old data.
            if (fully_expired.count(sst)) {
                continue;
            }

            // We also capture the sstable, so we keep it alive while the read isn't done
            ssts->insert(sst);
            // FIXME: If the sstables have cardinality estimation bitmaps, use that
            // for a better estimate for the number of partitions in the merged
            // sstable than just adding up the lengths of individual sstables.
            _estimated_partitions += sst->get_estimated_key_count();
            // TODO:
            // Note that this is not fully correct. Since we might be merging sstables that originated on
            // another shard (#cpu changed), we might be comparing RP:s with differing shard ids,
            // which might vary in "comparable" size quite a bit. However, since the worst that happens
            // is that we might miss a high water mark for the commit log replayer,
            // this is kind of ok, esp. since we will hopefully not be trying to recover based on
            // compacted sstables anyway (CL should be clean by then).
            _rp = std::max(_rp, sst->get_stats_metadata().position);
        }
        formatted_msg += "]";
        _info->sstables = _sstables.size();
        _info->ks = schema->ks_name();
        _info->cf = schema->cf_name();
        report_start(formatted_msg);

        return make_sstable_reader(std::move(ssts));
    }

    compaction_info finish(std::chrono::time_point<db_clock> started_at, std::chrono::time_point<db_clock> ended_at) {
        _info->ended_at = std::chrono::duration_cast<std::chrono::milliseconds>(ended_at.time_since_epoch()).count();
        auto ratio = double(_info->end_size) / double(_info->start_size);
        auto duration = std::chrono::duration<float>(ended_at - started_at);
        auto throughput = (double(_info->end_size) / (1024*1024)) / duration.count();
        sstring new_sstables_msg;

        for (auto& newtab : _info->new_sstables) {
            new_sstables_msg += sprint("%s:level=%d, ", newtab->get_filename(), newtab->get_sstable_level());
        }

        // FIXME: there is some missing information in the log message below.
        // look at CompactionTask::runMayThrow() in origin for reference.
        // - add support to merge summary (message: Partition merge counts were {%s}.).
        // - there is no easy way, currently, to know the exact number of total partitions.
        // By the time being, using estimated key count.
        sstring formatted_msg = sprint("%ld sstables to [%s]. %ld bytes to %ld (~%d%% of original) in %dms = %.2fMB/s. " \
            "~%ld total partitions merged to %ld.",
            _info->sstables, new_sstables_msg, _info->start_size, _info->end_size, int(ratio * 100),
            std::chrono::duration_cast<std::chrono::milliseconds>(duration).count(), throughput,
            _info->total_partitions, _info->total_keys_written);
        report_finish(formatted_msg, ended_at);

        backlog_tracker_adjust_charges();

        auto info = std::move(_info);
        _cf.get_compaction_manager().deregister_compaction(info);
        return std::move(*info);
    }

    virtual void report_start(const sstring& formatted_msg) const = 0;
    virtual void report_finish(const sstring& formatted_msg, std::chrono::time_point<db_clock> ended_at) const = 0;
    virtual void backlog_tracker_adjust_charges() = 0;

    virtual std::function<api::timestamp_type(const dht::decorated_key&)> max_purgeable_func() {
        return [] (const dht::decorated_key& dk) {
            return api::min_timestamp;
        };
    }

    virtual std::function<bool(const dht::decorated_key&)> filter_func() const {
        return [] (const dht::decorated_key&) {
            return true;
        };
    }

    // select a sstable writer based on decorated key.
    virtual sstable_writer* select_sstable_writer(const dht::decorated_key& dk) = 0;
    // stop current writer
    virtual void stop_sstable_writer() = 0;
    // finish all writers.
    virtual void finish_sstable_writer() = 0;

    compacting_sstable_writer get_compacting_sstable_writer() {
        return compacting_sstable_writer(*this);
    }

    const schema_ptr& schema() const {
        return _cf.schema();
    }
public:
    static future<compaction_info> run(std::unique_ptr<compaction> c);

    friend class compacting_sstable_writer;
};

void compacting_sstable_writer::consume_new_partition(const dht::decorated_key& dk) {
    if (_c._info->is_stop_requested()) {
        // Compaction manager will catch this exception and re-schedule the compaction.
        throw compaction_stop_exception(_c._info->ks, _c._info->cf, _c._info->stop_requested);
    }
    _writer = _c.select_sstable_writer(dk);
    _writer->consume_new_partition(dk);
    _c._info->total_keys_written++;
}

stop_iteration compacting_sstable_writer::consume_end_of_partition() {
    auto ret = _writer->consume_end_of_partition();
    if (ret == stop_iteration::yes) {
        // stop sstable writer being currently used.
        _c.stop_sstable_writer();
    }
    return ret;
}

void compacting_sstable_writer::consume_end_of_stream() {
    // this will stop any writer opened by compaction.
    _c.finish_sstable_writer();
}

class regular_compaction : public compaction {
    std::function<shared_sstable()> _creator;
    // store a clone of sstable set for column family, which needs to be alive for incremental selector.
    const sstable_set _set;
    // used to incrementally calculate max purgeable timestamp, as we iterate through decorated keys.
    sstable_set::incremental_selector _selector;
    // sstable being currently written.
    shared_sstable _sst;
    stdx::optional<sstable_writer> _writer;
    stdx::optional<compaction_weight_registration> _weight_registration;
    mutable compaction_read_monitor_generator _monitor_generator;
    std::deque<compaction_write_monitor> _active_write_monitors = {};
public:
    regular_compaction(column_family& cf, compaction_descriptor descriptor, std::function<shared_sstable()> creator,
            seastar::thread_scheduling_group* tsg)
        : compaction(cf, std::move(descriptor.sstables), descriptor.max_sstable_bytes, descriptor.level, tsg)
        , _creator(std::move(creator))
        , _set(cf.get_sstable_set())
        , _selector(_set.make_incremental_selector())
        , _weight_registration(std::move(descriptor.weight_registration))
        , _monitor_generator(_cf.get_compaction_manager(), _cf)
    {
    }

    flat_mutation_reader make_sstable_reader(lw_shared_ptr<sstables::sstable_set> ssts) const override {
        return ::make_local_shard_sstable_reader(_cf.schema(),
                std::move(ssts),
                query::full_partition_range,
                _cf.schema()->full_slice(),
                service::get_local_compaction_priority(),
                no_resource_tracking(),
                nullptr,
                ::streamed_mutation::forwarding::no,
                ::mutation_reader::forwarding::no,
                _monitor_generator);
    }

    void report_start(const sstring& formatted_msg) const override {
        clogger.info("Compacting {}", formatted_msg);
    }

    void report_finish(const sstring& formatted_msg, std::chrono::time_point<db_clock> ended_at) const override {
        clogger.info("Compacted {}", formatted_msg);
    }

    void backlog_tracker_adjust_charges() override {
        _monitor_generator.remove_sstables(_info->tracking);
        for (auto& wm : _active_write_monitors) {
            wm.add_sstable();
        }
    }

    virtual std::function<api::timestamp_type(const dht::decorated_key&)> max_purgeable_func() override {
        std::unordered_set<shared_sstable> compacting(_sstables.begin(), _sstables.end());
        return [this, compacting = std::move(compacting)] (const dht::decorated_key& dk) {
            return get_max_purgeable_timestamp(_cf, _selector, compacting, dk);
        };
    }

    virtual std::function<bool(const dht::decorated_key&)> filter_func() const override {
        return [] (const dht::decorated_key& dk){
            return dht::shard_of(dk.token()) == engine().cpu_id();
        };
    }

    virtual sstable_writer* select_sstable_writer(const dht::decorated_key& dk) override {
        if (!_writer) {
            _sst = _creator();
            setup_new_sstable(_sst);

            _active_write_monitors.emplace_back(_sst, _cf);
            auto&& priority = service::get_local_compaction_priority();
            sstable_writer_config cfg;
            cfg.max_sstable_size = _max_sstable_size;
            cfg.monitor = &_active_write_monitors.back();
            _writer.emplace(_sst->get_writer(*_cf.schema(), partitions_per_sstable(), cfg, priority));
        }
        return &*_writer;
    }

    virtual void stop_sstable_writer() override {
        finish_new_sstable(_writer, _sst);
    }

    virtual void finish_sstable_writer() override {
        on_end_of_stream();
        if (_writer) {
            stop_sstable_writer();
        }
    }
private:
    void on_end_of_stream() {
        if (_weight_registration) {
            _cf.get_compaction_manager().on_compaction_complete(*_weight_registration);
        }
    }
};

class cleanup_compaction final : public regular_compaction {
public:
    cleanup_compaction(column_family& cf, compaction_descriptor descriptor, std::function<shared_sstable()> creator,
            seastar::thread_scheduling_group* tsg)
        : regular_compaction(cf, std::move(descriptor), std::move(creator), tsg)
    {
        _info->type = compaction_type::Cleanup;
    }

    void report_start(const sstring& formatted_msg) const override {
        clogger.info("Cleaning {}", formatted_msg);
    }

    void report_finish(const sstring& formatted_msg, std::chrono::time_point<db_clock> ended_at) const override {
        clogger.info("Cleaned {}", formatted_msg);
    }

    std::function<bool(const dht::decorated_key&)> filter_func() const override {
        dht::token_range_vector owned_ranges = service::get_local_storage_service().get_local_ranges(_cf.schema()->ks_name());

        return [this, owned_ranges = std::move(owned_ranges)] (const dht::decorated_key& dk) {
            if (dht::shard_of(dk.token()) != engine().cpu_id()) {
                return false;
            }

            if (!belongs_to_current_node(dk.token(), owned_ranges)) {
                return false;
            }
            return true;
        };
    }
};


class resharding_compaction final : public compaction {
    std::vector<std::pair<shared_sstable, stdx::optional<sstable_writer>>> _output_sstables;
    shard_id _shard; // shard of current sstable writer
    std::function<shared_sstable(shard_id)> _sstable_creator;
    compaction_backlog_tracker _resharding_backlog_tracker;
public:
    resharding_compaction(std::vector<shared_sstable> sstables, column_family& cf, std::function<shared_sstable(shard_id)> creator,
            uint64_t max_sstable_size, uint32_t sstable_level, seastar::thread_scheduling_group* tsg)
        : compaction(cf, std::move(sstables), max_sstable_size, sstable_level, tsg)
        , _output_sstables(smp::count)
        , _sstable_creator(std::move(creator))
        , _resharding_backlog_tracker(std::make_unique<resharding_backlog_tracker>())
    {
        cf.get_compaction_manager().register_backlog_tracker(_resharding_backlog_tracker);
        for (auto& s : _sstables) {
            _resharding_backlog_tracker.add_sstable(s);
        }
        _info->type = compaction_type::Reshard;
    }

    ~resharding_compaction() {
        for (auto& s : _sstables) {
            _resharding_backlog_tracker.remove_sstable(s);
        }
    }

    // Use reader that makes sure no non-local mutation will not be filtered out.
    flat_mutation_reader make_sstable_reader(lw_shared_ptr<sstables::sstable_set> ssts) const override {
        return ::make_range_sstable_reader(_cf.schema(),
                std::move(ssts),
                query::full_partition_range,
                _cf.schema()->full_slice(),
                service::get_local_compaction_priority(),
                no_resource_tracking(),
                nullptr,
                ::streamed_mutation::forwarding::no,
                ::mutation_reader::forwarding::no);
    }

    void report_start(const sstring& formatted_msg) const override {
        clogger.info("Resharding {}", formatted_msg);
    }

    void report_finish(const sstring& formatted_msg, std::chrono::time_point<db_clock> ended_at) const override {
        clogger.info("Resharded {}", formatted_msg);
    }

    void backlog_tracker_adjust_charges() override { }

    sstable_writer* select_sstable_writer(const dht::decorated_key& dk) override {
        _shard = dht::shard_of(dk.token());
        auto& sst = _output_sstables[_shard].first;
        auto& writer = _output_sstables[_shard].second;

        if (!writer) {
            sst = _sstable_creator(_shard);
            setup_new_sstable(sst);

            sstable_writer_config cfg;
            cfg.max_sstable_size = _max_sstable_size;
            auto&& priority = service::get_local_compaction_priority();
            writer.emplace(sst->get_writer(*_cf.schema(), partitions_per_sstable(), cfg, priority, _shard));
        }
        return &*writer;
    }

    void stop_sstable_writer() override {
        auto& sst = _output_sstables[_shard].first;
        auto& writer = _output_sstables[_shard].second;

        finish_new_sstable(writer, sst);
    }

    void finish_sstable_writer() override {
        for (auto& p : _output_sstables) {
            if (p.second) {
                finish_new_sstable(p.second, p.first);
            }
        }
    }
};

future<compaction_info> compaction::run(std::unique_ptr<compaction> c) {
    auto attr = c->thread_attributes();
    return seastar::async(std::move(attr), [c = std::move(c)] () mutable {
        auto reader = c->setup();

        auto cr = c->get_compacting_sstable_writer();
        auto cfc = make_stable_flattened_mutations_consumer<compact_for_compaction<compacting_sstable_writer>>(
            *c->schema(), gc_clock::now(), std::move(cr), c->max_purgeable_func());

        auto start_time = db_clock::now();
        try {
            // make sure the readers are all gone before the compaction object is gone. We will
            // leave this block either successfully or exceptionally with the reader object
            // destroyed.
            auto r = std::move(reader);
            r.consume_in_thread(std::move(cfc), c->filter_func());
        } catch (...) {
            delete_sstables_for_interrupted_compaction(c->_info->new_sstables, c->_info->ks, c->_info->cf);
            c = nullptr; // make sure writers are stopped while running in thread context
            throw;
        }

        return c->finish(std::move(start_time), db_clock::now());
    });
}

template <typename ...Params>
static std::unique_ptr<compaction> make_compaction(bool cleanup, Params&&... params) {
    if (cleanup) {
        return std::make_unique<cleanup_compaction>(std::forward<Params>(params)...);
    } else {
        return std::make_unique<regular_compaction>(std::forward<Params>(params)...);
    }
}

future<compaction_info>
compact_sstables(sstables::compaction_descriptor descriptor, column_family& cf, std::function<shared_sstable()> creator,
        bool cleanup, seastar::thread_scheduling_group *tsg) {
    if (descriptor.sstables.empty()) {
        throw std::runtime_error(sprint("Called compaction with empty set on behalf of {}.{}", cf.schema()->ks_name(), cf.schema()->cf_name()));
    }
    auto c = make_compaction(cleanup, cf, std::move(descriptor), std::move(creator), tsg);
    return compaction::run(std::move(c));
}

future<std::vector<shared_sstable>>
reshard_sstables(std::vector<shared_sstable> sstables, column_family& cf, std::function<shared_sstable(shard_id)> creator,
        uint64_t max_sstable_size, uint32_t sstable_level, seastar::thread_scheduling_group* tsg) {
    if (sstables.empty()) {
        throw std::runtime_error(sprint("Called resharding with empty set on behalf of {}.{}", cf.schema()->ks_name(), cf.schema()->cf_name()));
    }
    auto c = std::make_unique<resharding_compaction>(std::move(sstables), cf, std::move(creator), max_sstable_size, sstable_level, tsg);
    return compaction::run(std::move(c)).then([] (auto ret) {
        return std::move(ret.new_sstables);
    });
}

std::unordered_set<sstables::shared_sstable>
get_fully_expired_sstables(column_family& cf, const std::vector<sstables::shared_sstable>& compacting, gc_clock::time_point gc_before) {
    clogger.debug("Checking droppable sstables in {}.{}", cf.schema()->ks_name(), cf.schema()->cf_name());

    if (compacting.empty()) {
        return {};
    }

    std::unordered_set<sstables::shared_sstable> candidates;
    auto uncompacting_sstables = get_uncompacting_sstables(cf, compacting);
    // Get list of uncompacting sstables that overlap the ones being compacted.
    std::vector<sstables::shared_sstable> overlapping = leveled_manifest::overlapping(*cf.schema(), compacting, uncompacting_sstables);
    int64_t min_timestamp = std::numeric_limits<int64_t>::max();

    for (auto& sstable : overlapping) {
        if (sstable->get_max_local_deletion_time() >= gc_before) {
            min_timestamp = std::min(min_timestamp, sstable->get_stats_metadata().min_timestamp);
        }
    }

    auto compacted_undeleted_gens = boost::copy_range<std::unordered_set<int64_t>>(cf.compacted_undeleted_sstables()
        | boost::adaptors::transformed(std::mem_fn(&sstables::sstable::generation)));
    auto has_undeleted_ancestor = [&compacted_undeleted_gens] (auto& candidate) {
        return boost::algorithm::any_of(candidate->ancestors(), [&compacted_undeleted_gens] (auto gen) {
            return compacted_undeleted_gens.count(gen);
        });
    };

    // SStables that do not contain live data is added to list of possibly expired sstables.
    for (auto& candidate : compacting) {
        clogger.debug("Checking if candidate of generation {} and max_deletion_time {} is expired, gc_before is {}",
                    candidate->generation(), candidate->get_stats_metadata().max_local_deletion_time, gc_before);
        // A fully expired sstable which has an ancestor undeleted shouldn't be compacted because
        // expired data won't be purged because undeleted sstables are taken into account when
        // calculating max purgeable timestamp, and not doing it could lead to a compaction loop.
        if (candidate->get_max_local_deletion_time() < gc_before && !has_undeleted_ancestor(candidate)) {
            clogger.debug("Adding candidate of generation {} to list of possibly expired sstables", candidate->generation());
            candidates.insert(candidate);
        } else {
            min_timestamp = std::min(min_timestamp, candidate->get_stats_metadata().min_timestamp);
        }
    }

    auto it = candidates.begin();
    while (it != candidates.end()) {
        auto& candidate = *it;
        // Remove from list any candidate that may contain a tombstone that covers older data.
        if (candidate->get_stats_metadata().max_timestamp >= min_timestamp) {
            it = candidates.erase(it);
        } else {
            clogger.debug("Dropping expired SSTable {} (maxLocalDeletionTime={}, gcBefore={})",
                    candidate->get_filename(), candidate->get_stats_metadata().max_local_deletion_time, gc_before);
            it++;
        }
    }
    return candidates;
}

}