Try to start TTL earlier with kMinOverlappingRatio is used (#8749)
Summary: Right now, when options.ttl is set, compactions are triggered around the time when TTL is reached. This might cause extra compactions which are often bursty. This commit tries to mitigate it by picking those files earlier in normal compaction picking process. This is only implemented using kMinOverlappingRatio with Leveled compaction as it is the default value and it is more complicated to change other styles. When a file is aged more than ttl/2, RocksDB starts to boost the compaction priority of files in normal compaction picking process, and hope by the time TTL is reached, very few extra compaction is needed. In order for this to work, another change is made: during a compaction, if an output level file is older than ttl/2, cut output files based on original boundary (if it is not in the last level). This is to make sure that after an old file is moved to the next level, and new data is merged from the upper level, the new data falling into this range isn't reset with old timestamp. Without this change, in many cases, most files from one level will keep having old timestamp, even if they have newer data and we stuck in it. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8749 Test Plan: Add a unit test to test the boosting logic. Will add a unit test to test it end-to-end. Reviewed By: jay-zhuang Differential Revision: D30735261 fbshipit-source-id: 503c2d89250b22911eb99e72b379be154de3428emain
sdong
3 years ago
committed by
Facebook GitHub Bot
parent
a5ec5e3ea0
commit
a2b9be42b6
@ -0,0 +1,92 @@ |
||||
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
||||
// This source code is licensed under both the GPLv2 (found in the
|
||||
// COPYING file in the root directory) and Apache 2.0 License
|
||||
// (found in the LICENSE.Apache file in the root directory).
|
||||
//
|
||||
|
||||
#pragma once |
||||
#include <algorithm> |
||||
|
||||
#include "db/version_edit.h" |
||||
|
||||
namespace ROCKSDB_NAMESPACE { |
||||
// We boost files that are closer to TTL limit. This boosting could be
|
||||
// through FileMetaData.compensated_file_size but this compensated size
|
||||
// is widely used as something similar to file size so dramatically boost
|
||||
// the value might cause unintended consequences.
|
||||
//
|
||||
// This boosting algorithm can go very fancy, but here we use a simple
|
||||
// formula which can satisify:
|
||||
// (1) Different levels are triggered slightly differently to avoid
|
||||
// too many cascading cases
|
||||
// (2) Files in the same level get boosting more when TTL gets closer.
|
||||
//
|
||||
// Don't do any boosting before TTL has past by half. This is to make
|
||||
// sure lower write amp for most of the case. And all levels should be
|
||||
// fully boosted when total TTL compaction threshold triggers.
|
||||
// Differientiate boosting ranges of each level by 1/2. This will make
|
||||
// range for each level exponentially increasing. We could do it by
|
||||
// having them to be equal, or go even fancier. We can adjust it after
|
||||
// we observe the behavior in production.
|
||||
// The threshold starting boosting:
|
||||
// +------------------------------------------------------------------ +
|
||||
// ^ ^ ^ ^ ^ ^
|
||||
// Age 0 ... | | second last level thresold
|
||||
// | |
|
||||
// | third last level
|
||||
// |
|
||||
// forth last level
|
||||
//
|
||||
// We arbitrarily set with 0 when a file is aged boost_age_start and
|
||||
// grow linearly. The ratio is arbitrarily set so that when the next level
|
||||
// starts to boost, the previous level's boosting amount is 16.
|
||||
class FileTtlBooster { |
||||
public: |
||||
FileTtlBooster(uint64_t current_time, uint64_t ttl, int num_non_empty_levels, |
||||
int level) |
||||
: current_time_(current_time) { |
||||
if (ttl == 0 || level == 0 || level >= num_non_empty_levels - 1) { |
||||
enabled_ = false; |
||||
boost_age_start_ = 0; |
||||
boost_step_ = 1; |
||||
} else { |
||||
enabled_ = true; |
||||
uint64_t all_boost_start_age = ttl / 2; |
||||
uint64_t all_boost_age_range = (ttl / 32) * 31 - all_boost_start_age; |
||||
uint64_t boost_age_range = |
||||
all_boost_age_range >> (num_non_empty_levels - level - 1); |
||||
boost_age_start_ = all_boost_start_age + boost_age_range; |
||||
const uint64_t kBoostRatio = 16; |
||||
// prevent 0 value to avoid divide 0 error.
|
||||
boost_step_ = std::max(boost_age_range / kBoostRatio, uint64_t{1}); |
||||
} |
||||
} |
||||
|
||||
uint64_t GetBoostScore(FileMetaData* f) { |
||||
if (!enabled_) { |
||||
return 1; |
||||
} |
||||
uint64_t oldest_ancester_time = f->TryGetOldestAncesterTime(); |
||||
if (oldest_ancester_time >= current_time_) { |
||||
return 1; |
||||
} |
||||
uint64_t age = current_time_ - oldest_ancester_time; |
||||
if (age > boost_age_start_) { |
||||
// Use integer just for convenience.
|
||||
// We could make all file_to_order double if we want.
|
||||
// Technically this can overflow if users override timing and
|
||||
// give a very high current time. Ignore the case for simplicity.
|
||||
// Boosting is addition to current value, so +1. This will effectively
|
||||
// make boosting to kick in after the first boost_step_ is reached.
|
||||
return (age - boost_age_start_) / boost_step_ + 1; |
||||
} |
||||
return 1; |
||||
} |
||||
|
||||
private: |
||||
bool enabled_; |
||||
uint64_t current_time_; |
||||
uint64_t boost_age_start_; |
||||
uint64_t boost_step_; |
||||
}; |
||||
} // namespace ROCKSDB_NAMESPACE
|
||||
Loading…
Reference in new issue