1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
|
// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
// vim: ts=8 sw=2 smarttab
/*
* Ceph - scalable distributed file system
*
* Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
*
* This is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License version 2.1, as published by the Free Software
* Foundation. See file COPYING.
*
*/
#ifndef CEPH_MDBALANCER_H
#define CEPH_MDBALANCER_H
#include "include/types.h"
#include "common/Clock.h"
#include "common/Cond.h"
#include "msg/Message.h"
#include "messages/MHeartbeat.h"
#include "MDSMap.h"
class MDSRank;
class MHeartbeat;
class CInode;
class CDir;
class Messenger;
class MonClient;
class MDBalancer {
public:
using clock = ceph::coarse_mono_clock;
using time = ceph::coarse_mono_time;
friend class C_Bal_SendHeartbeat;
MDBalancer(MDSRank *m, Messenger *msgr, MonClient *monc);
void handle_conf_change(const std::set<std::string>& changed, const MDSMap& mds_map);
int proc_message(const cref_t<Message> &m);
/**
* Regularly called upkeep function.
*
* Sends MHeartbeat messages to the mons.
*/
void tick();
void handle_export_pins(void);
void subtract_export(CDir *ex);
void add_import(CDir *im);
void adjust_pop_for_rename(CDir *pdir, CDir *dir, bool inc);
void hit_inode(CInode *in, int type, int who=-1);
void hit_dir(CDir *dir, int type, int who=-1, double amount=1.0);
void queue_split(const CDir *dir, bool fast);
void queue_merge(CDir *dir);
bool is_fragment_pending(dirfrag_t df) {
return split_pending.count(df) || merge_pending.count(df);
}
/**
* Based on size and configuration, decide whether to issue a queue_split
* or queue_merge for this CDir.
*
* \param hot whether the directory's temperature is enough to split it
*/
void maybe_fragment(CDir *dir, bool hot);
void handle_mds_failure(mds_rank_t who);
int dump_loads(Formatter *f) const;
private:
typedef struct {
std::map<mds_rank_t, double> targets;
std::map<mds_rank_t, double> imported;
std::map<mds_rank_t, double> exported;
} balance_state_t;
//set up the rebalancing targets for export and do one if the
//MDSMap is up to date
void prep_rebalance(int beat);
int mantle_prep_rebalance();
mds_load_t get_load();
int localize_balancer();
void send_heartbeat();
void handle_heartbeat(const cref_t<MHeartbeat> &m);
void find_exports(CDir *dir,
double amount,
std::vector<CDir*>* exports,
double& have,
set<CDir*>& already_exporting);
double try_match(balance_state_t &state,
mds_rank_t ex, double& maxex,
mds_rank_t im, double& maxim);
double get_maxim(balance_state_t &state, mds_rank_t im) {
return target_load - mds_meta_load[im] - state.imported[im];
}
double get_maxex(balance_state_t &state, mds_rank_t ex) {
return mds_meta_load[ex] - target_load - state.exported[ex];
}
/**
* Try to rebalance.
*
* Check if the monitor has recorded the current export targets;
* if it has then do the actual export. Otherwise send off our
* export targets message again.
*/
void try_rebalance(balance_state_t& state);
bool bal_fragment_dirs;
int64_t bal_fragment_interval;
static const unsigned int AUTH_TREES_THRESHOLD = 5;
MDSRank *mds;
Messenger *messenger;
MonClient *mon_client;
int beat_epoch = 0;
string bal_code;
string bal_version;
time last_heartbeat = clock::zero();
time last_sample = clock::zero();
time rebalance_time = clock::zero(); //ensure a consistent view of load for rebalance
time last_get_load = clock::zero();
uint64_t last_num_requests = 0;
uint64_t last_cpu_time = 0;
uint64_t last_num_traverse = 0;
uint64_t last_num_traverse_hit = 0;
// Dirfrags which are marked to be passed on to MDCache::[split|merge]_dir
// just as soon as a delayed context comes back and triggers it.
// These sets just prevent us from spawning extra timer contexts for
// dirfrags that already have one in flight.
set<dirfrag_t> split_pending, merge_pending;
// per-epoch scatter/gathered info
std::map<mds_rank_t, mds_load_t> mds_load;
std::map<mds_rank_t, double> mds_meta_load;
std::map<mds_rank_t, map<mds_rank_t, float> > mds_import_map;
std::map<mds_rank_t, int> mds_last_epoch_under_map;
// per-epoch state
double my_load = 0;
double target_load = 0;
};
#endif
|
