Move LPOptimizer, DSL and tests to lp_optimizer/

This commit is contained in:
Vitaliy Filippov
2024-06-05 10:51:20 +03:00
parent 86832dc43f
commit 848c2d2722
13 changed files with 10 additions and 10 deletions
+409
View File
@@ -0,0 +1,409 @@
const { select_murmur3 } = require('./murmur3.js');
const NO_OSD = 'Z';
class RuleCombinator
{
constructor(osd_tree, rules, max_combinations, ordered)
{
this.osd_tree = index_tree(Object.values(osd_tree).filter(o => o.id));
this.rules = rules;
this.max_combinations = max_combinations;
this.ordered = ordered;
}
random_combinations()
{
return random_custom_combinations(this.osd_tree, this.rules, this.max_combinations, this.ordered);
}
check_combinations(pgs)
{
return check_custom_combinations(this.osd_tree, this.rules, pgs);
}
}
// Convert alternative "level-index" format to rules
// level_index = { [level: string]: string | string[] }
// level_sequence = optional, levels from upper to lower, i.e. [ 'dc', 'host' ]
// Example: level_index = { dc: "112233", host: "ABCDEF" }
function parse_level_indexes(level_index, level_sequence)
{
const rules = [];
const lvl_first = {};
for (const level in level_index)
{
const idx = level_index[level];
while (rules.length < idx.length)
{
rules.push([]);
}
const seen = {};
for (let i = 0; i < idx.length; i++)
{
if (!seen[idx[i]])
{
const other = Object.values(seen);
if (other.length)
{
rules[i].push([ level, '!=', other ]);
}
seen[idx[i]] = i+1;
}
else
{
rules[i].push([ level, '=', seen[idx[i]] ]);
}
}
lvl_first[level] = seen;
}
if (level_sequence)
{
// Prune useless rules for the sake of prettiness
// For simplicity, call "upper" level DC and "lower" level host
const level_prio = Object.keys(level_sequence).reduce((a, c) => { a[level_sequence[c]] = c; return a; }, {});
for (let upper_i = 0; upper_i < level_sequence.length-1; upper_i++)
{
const upper_level = level_sequence[upper_i];
for (let i = 0; i < rules.length; i++)
{
const noteq = {};
for (let k = 0; k < level_index[upper_level].length; k++)
{
// If upper_level[x] is different from upper_level[y]
// then lower_level[x] is also different from lower_level[y]
if (level_index[upper_level][k] != level_index[upper_level][i])
{
noteq[k+1] = true;
}
}
for (let j = 0; j < rules[i].length; j++)
{
if (level_prio[rules[i][j][0]] != null && level_prio[rules[i][j][0]] > upper_i && rules[i][j][1] == '!=')
{
rules[i][j][2] = rules[i][j][2].filter(other_host => !noteq[other_host]);
if (!rules[i][j][2].length)
{
rules[i].splice(j--, 1);
}
}
}
}
}
}
return rules;
}
// Parse rules in DSL format
// dsl := item | item ("\n" | ",") items
// item := "any" | rules
// rules := rule | rule rules
// rule := level operator arg
// level := /\w+/
// operator := "!=" | "=" | ">" | "?="
// arg := value | "(" values ")"
// values := value | value "," values
// value := item_ref | constant_id
// item_ref := /\d+/
// constant_id := /"([^"]+)"/
//
// Output: [ level, operator, value ][][]
function parse_pg_dsl(text)
{
const tokens = [ ...text.matchAll(/\w+|!=|\?=|[>=\(\),\n]|"([^\"]+)"/g) ].map(t => [ t[0], t.index ]);
let positions = [ [] ];
let rules = positions[0];
for (let i = 0; i < tokens.length; )
{
if (tokens[i][0] === '\n' || tokens[i][0] === ',')
{
rules = [];
positions.push(rules);
i++;
}
else if (!rules.length && tokens[i][0] === 'any' && (i == tokens.length-1 || tokens[i+1][0] === ',' || tokens[i+1][0] === '\n'))
{
i++;
}
else
{
if (!/^\w/.exec(tokens[i][0]))
{
throw new Error('Unexpected '+tokens[i][0]+' at '+tokens[i][1]+' (level name expected)');
}
if (i > tokens.length-3)
{
throw new Error('Unexpected EOF (operator and value expected)');
}
if (/^\w/.exec(tokens[i+1][0]) || tokens[i+1][0] === ',' || tokens[i+1][0] === '\n')
{
throw new Error('Unexpected '+tokens[i+1][0]+' at '+tokens[i+1][1]+' (operator expected)');
}
if (!/^[\w"(]/.exec(tokens[i+2][0])) // "
{
throw new Error('Unexpected '+tokens[i+2][0]+' at '+tokens[i+2][1]+' (id, round brace, number or node ID expected)');
}
let rule = [ tokens[i][0], tokens[i+1][0], tokens[i+2][0] ];
i += 3;
if (rule[2][0] == '"')
{
rule[2] = { id: rule[2].substr(1, rule[2].length-2) };
}
else if (rule[2] === '(')
{
rule[2] = [];
// eslint-disable-next-line no-constant-condition
while (true)
{
if (i > tokens.length-1)
{
throw new Error('Unexpected EOF (expected list and a closing round brace)');
}
if (tokens[i][0] === ',')
{
i++;
}
else if (tokens[i][0] === ')')
{
i++;
break;
}
else if (tokens[i][0][0] === '"')
{
rule[2].push({ id: tokens[i][0].substr(1, tokens[i][0].length-2) });
i++;
}
else if (/^\d+$/.exec(tokens[i][0]))
{
const n = 0|tokens[i][0];
if (!n)
{
throw new Error('Level reference cannot be 0 (refs count from 1) at '+tokens[i][1]);
}
else if (n > positions.length)
{
throw new Error('Forward references are forbidden at '+tokens[i][1]);
}
rule[2].push(n);
i++;
}
else if (!/^\w/.exec(tokens[i][0]))
{
throw new Error('Unexpected '+tokens[i][0]+' at '+tokens[i][1]+' (number or node ID expected)');
}
else
{
rule[2].push({ id: tokens[i][0] });
i++;
}
}
}
else if (!/^\d+$/.exec(rule[2]))
{
rule[2] = { id: rule[2] };
}
else
{
rule[2] = 0|rule[2];
if (!rule[2])
{
throw new Error('Level reference cannot be 0 (refs count from 1) at '+tokens[i-1][1]);
}
else if (rule[2] > positions.length)
{
throw new Error('Forward references are forbidden at '+tokens[i-1][1]);
}
}
rules.push(rule);
}
}
return positions;
}
// osd_tree = index_tree() output
// levels = { string: number }
// rules = [ level, operator, value ][][]
// level = string
// operator = '=' | '!=' | '>' | '?='
// value = number|number[] | { id: string|string[] }
// examples:
// 1) simple 3 replicas with failure_domain=host:
// [ [], [ [ 'host', '!=', 1 ] ], [ [ 'host', '!=', [ 1, 2 ] ] ] ]
// in DSL form: any, host!=1, host!=(1,2)
// 2) EC 4+2 in 3 DC:
// [ [], [ [ 'dc', '=', 1 ], [ 'host', '!=', 1 ] ],
// [ 'dc', '!=', 1 ], [ [ 'dc', '=', 3 ], [ 'host', '!=', 3 ] ],
// [ 'dc', '!=', [ 1, 3 ] ], [ [ 'dc', '=', 5 ], [ 'host', '!=', 5 ] ] ]
// in DSL form: any, dc=1 host!=1, dc!=1, dc=3 host!=3, dc!=(1,3), dc=5 host!=5
// 3) 1 replica in fixed DC + 2 in random DCs:
// [ [ [ 'dc', '=', { id: 'meow' } ] ], [ [ 'dc', '!=', 1 ] ], [ [ 'dc', '!=', [ 1, 2 ] ] ] ]
// in DSL form: dc=meow, dc!=1, dc!=(1,2)
// 4) 2 replicas in each DC (almost the same as (2)):
// DSL: any, dc=1 host!=1, dc!=1, dc=3 host!=3
// Alternative simpler way to specify rules would be: [ DC: 112233 HOST: 123456 ]
function random_custom_combinations(osd_tree, rules, count, ordered)
{
const r = {};
const first = filter_tree_by_rules(osd_tree, rules[0], []);
let max_size = 0;
// All combinations for the first item (usually "any") to try to include each OSD at least once
for (const f of first)
{
const selected = [ f ];
for (let i = 1; i < rules.length; i++)
{
const filtered = filter_tree_by_rules(osd_tree, rules[i], selected);
const idx = select_murmur3(filtered.length, i => 'p:'+f.id+':'+filtered[i].id);
selected.push(idx == null ? { levels: {}, id: null } : filtered[idx]);
}
const size = selected.filter(s => s.id !== null).length;
max_size = max_size < size ? size : max_size;
const pg = selected.map(s => s.id === null ? NO_OSD : (0|s.id));
if (!ordered)
pg.sort();
r['pg_'+pg.join('_')] = pg;
}
// Pseudo-random selection
for (let n = 0; n < count; n++)
{
const selected = [];
for (const item_rules of rules)
{
const filtered = selected.length ? filter_tree_by_rules(osd_tree, item_rules, selected) : first;
const idx = select_murmur3(filtered.length, i => n+':'+filtered[i].id);
selected.push(idx == null ? { levels: {}, id: null } : filtered[idx]);
}
const size = selected.filter(s => s.id !== null).length;
max_size = max_size < size ? size : max_size;
const pg = selected.map(s => s.id === null ? NO_OSD : (0|s.id));
if (!ordered)
pg.sort();
r['pg_'+pg.join('_')] = pg;
}
// Exclude PGs with less successful selections than maximum
for (const k in r)
{
if (r[k].filter(s => s !== NO_OSD).length < max_size)
{
delete r[k];
}
}
return r;
}
function filter_tree_by_rules(osd_tree, rules, selected)
{
let cur = osd_tree[''].children;
for (const rule of rules)
{
const val = (rule[2] instanceof Array ? rule[2] : [ rule[2] ])
.map(v => v instanceof Object ? v.id : selected[v-1].levels[rule[0]]);
let preferred = [], other = [];
for (let i = 0; i < cur.length; i++)
{
const item = cur[i];
const level_id = item.levels[rule[0]];
if (level_id)
{
if (rule[1] == '>' && val.filter(v => level_id <= v).length == 0 ||
(rule[1] == '=' || rule[1] == '?=') && val.filter(v => level_id != v).length == 0 ||
rule[1] == '!=' && val.filter(v => level_id == v).length == 0)
{
// Include
preferred.push(item);
}
else if (rule[1] == '?=' && val.filter(v => level_id != v).length > 0)
{
// Non-preferred
other.push(item);
}
}
else if (item.children)
{
// Descend
cur.splice(i+1, 0, ...item.children);
}
}
cur = preferred.length ? preferred : other;
}
// Get leaf items
for (let i = 0; i < cur.length; i++)
{
if (cur[i].children)
{
// Descend
cur.splice(i, 1, ...cur[i].children);
i--;
}
}
return cur;
}
// Convert from
// node_list = { id: string|number, level: string, size?: number, parent?: string|number }[]
// to
// node_tree = { [node_id]: { id, level, size?, parent?, children?: child_node_id[], levels: { [level]: id, ... } } }
function index_tree(node_list)
{
const tree = { '': { children: [], levels: {} } };
for (const node of node_list)
{
tree[node.id] = { ...node, levels: {} };
delete tree[node.id].children;
}
for (const node of node_list)
{
const parent_id = node.parent && tree[node.parent] ? node.parent : '';
tree[parent_id].children = tree[parent_id].children || [];
tree[parent_id].children.push(tree[node.id]);
}
const cur = tree[''].children;
for (let i = 0; i < cur.length; i++)
{
cur[i].levels[cur[i].level] = cur[i].id;
if (cur[i].children)
{
for (const child of cur[i].children)
{
child.levels = { ...cur[i].levels, ...child.levels };
}
cur.splice(i, 1, ...cur[i].children);
i--;
}
}
return tree;
}
// selection = id[]
// osd_tree = index_tree output
// rules = parse_pg_dsl output
function check_custom_combinations(osd_tree, rules, pgs)
{
const res = [];
skip_pg: for (const pg of pgs)
{
let selected = pg.map(id => osd_tree[id] || null);
for (let i = 0; i < rules.length; i++)
{
const filtered = filter_tree_by_rules(osd_tree, rules[i], selected);
if (selected[i] === null && filtered.length ||
!filtered.filter(ok => selected[i].id === ok.id).length)
{
continue skip_pg;
}
}
res.push(pg);
}
return res;
}
module.exports = {
RuleCombinator,
NO_OSD,
index_tree,
parse_level_indexes,
parse_pg_dsl,
random_custom_combinations,
check_custom_combinations,
};
+555
View File
@@ -0,0 +1,555 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
// Data distribution optimizer using linear programming (lp_solve)
const child_process = require('child_process');
const NO_OSD = 'Z';
async function lp_solve(text)
{
const cp = child_process.spawn('lp_solve');
let stdout = '', stderr = '', finish_cb;
cp.stdout.on('data', buf => stdout += buf.toString());
cp.stderr.on('data', buf => stderr += buf.toString());
cp.on('exit', () => finish_cb && finish_cb());
cp.stdin.write(text);
cp.stdin.end();
if (cp.exitCode == null)
{
await new Promise(ok => finish_cb = ok);
}
if (!stdout.trim())
{
return null;
}
let score = 0;
let vars = {};
for (const line of stdout.split(/\n/))
{
let m = /^(^Value of objective function: (-?[\d\.]+)|Actual values of the variables:)\s*$/.exec(line);
if (m)
{
if (m[2])
{
score = m[2];
}
continue;
}
else if (/This problem is (infeasible|unbounded)/.exec(line))
{
return null;
}
let [ k, v ] = line.trim().split(/\s+/, 2);
if (v)
{
vars[k] = v;
}
}
return { score, vars };
}
// osd_weights = { [id]: weight }
async function optimize_initial({ osd_weights, combinator, pg_count, pg_size = 3, pg_minsize = 2, parity_space = 1, ordered = false })
{
if (!pg_count || !osd_weights)
{
return null;
}
const total_weight = Object.values(osd_weights).reduce((a, c) => Number(a) + Number(c), 0);
const all_pgs = Object.values(make_cyclic(combinator.random_combinations(), parity_space));
const pg_per_osd = {};
for (const pg of all_pgs)
{
for (let i = 0; i < pg.length; i++)
{
const osd = pg[i];
pg_per_osd[osd] = pg_per_osd[osd] || [];
pg_per_osd[osd].push((i >= pg_minsize ? parity_space+'*' : '')+"pg_"+pg.join("_"));
}
}
let pg_effsize = all_pgs.reduce((a, c) => Math.max(a, c.filter(e => e != NO_OSD).length), 0);
pg_effsize = Math.min(pg_minsize, pg_effsize) + Math.max(0, Math.min(pg_size, pg_effsize) - pg_minsize) * parity_space;
let lp = '';
lp += "max: "+all_pgs.map(pg => 'pg_'+pg.join('_')).join(' + ')+";\n";
for (const osd in pg_per_osd)
{
if (osd !== NO_OSD)
{
let osd_pg_count = osd_weights[osd]/total_weight*pg_effsize*pg_count;
lp += pg_per_osd[osd].join(' + ')+' <= '+osd_pg_count+';\n';
}
}
for (const pg of all_pgs)
{
lp += 'pg_'+pg.join('_')+" >= 0;\n";
}
lp += "sec "+all_pgs.map(pg => 'pg_'+pg.join('_')).join(', ')+";\n";
const lp_result = await lp_solve(lp);
if (!lp_result)
{
console.log(lp);
throw new Error('Problem is infeasible or unbounded - is it a bug?');
}
const int_pgs = make_int_pgs(lp_result.vars, pg_count, ordered);
const eff = pg_list_space_efficiency(int_pgs, osd_weights, pg_minsize, parity_space);
const res = {
score: lp_result.score,
weights: lp_result.vars,
int_pgs,
space: eff * pg_effsize,
total_space: total_weight,
};
return res;
}
function make_cyclic(pgs, parity_space)
{
if (parity_space > 1)
{
for (const pg in pgs)
{
for (let i = 1; i < pg.size; i++)
{
const cyclic = [ ...pg.slice(i), ...pg.slice(0, i) ];
pgs['pg_'+cyclic.join('_')] = cyclic;
}
}
}
return pgs;
}
function shuffle(array)
{
for (let i = array.length - 1, j, x; i > 0; i--)
{
j = Math.floor(Math.random() * (i + 1));
x = array[i];
array[i] = array[j];
array[j] = x;
}
}
function make_int_pgs(weights, pg_count, round_robin)
{
const total_weight = Object.values(weights).reduce((a, c) => Number(a) + Number(c), 0);
let int_pgs = [];
let pg_left = pg_count;
let weight_left = total_weight;
for (const pg_name in weights)
{
let cur_pg = pg_name.substr(3).split('_');
let n = Math.round(weights[pg_name] / weight_left * pg_left);
for (let i = 0; i < n; i++)
{
int_pgs.push([ ...cur_pg ]);
if (round_robin)
{
cur_pg.push(cur_pg.shift());
}
}
weight_left -= weights[pg_name];
pg_left -= n;
}
shuffle(int_pgs);
return int_pgs;
}
function calc_intersect_weights(old_pg_size, pg_size, pg_count, prev_weights, all_pgs, ordered)
{
const move_weights = {};
if ((1 << old_pg_size) < pg_count)
{
const intersect = {};
for (const pg_name in prev_weights)
{
const pg = pg_name.substr(3).split(/_/);
for (let omit = 1; omit < (1 << old_pg_size); omit++)
{
let pg_omit = [ ...pg ];
let intersect_count = old_pg_size;
for (let i = 0; i < old_pg_size; i++)
{
if (omit & (1 << i))
{
pg_omit[i] = '';
intersect_count--;
}
}
if (!ordered)
pg_omit = pg_omit.filter(n => n).sort();
pg_omit = pg_omit.join(':');
intersect[pg_omit] = Math.max(intersect[pg_omit] || 0, intersect_count);
}
}
for (const pg of all_pgs)
{
let max_int = 0;
for (let omit = 1; omit < (1 << pg_size); omit++)
{
let pg_omit = [ ...pg ];
for (let i = 0; i < pg_size; i++)
{
if (omit & (1 << i))
pg_omit[i] = '';
}
if (!ordered)
pg_omit = pg_omit.filter(n => n).sort();
pg_omit = pg_omit.join(':');
max_int = Math.max(max_int, intersect[pg_omit] || 0);
}
move_weights['pg_'+pg.join('_')] = pg_size-max_int;
}
}
else
{
const prev_pg_hashed = Object.keys(prev_weights).map(pg_name => pg_name
.substr(3).split(/_/).reduce((a, c, i) => { a[c] = i+1; return a; }, {}));
for (const pg of all_pgs)
{
if (!prev_weights['pg_'+pg.join('_')])
{
let max_int = 0;
for (const prev_hash of prev_pg_hashed)
{
const intersect_count = ordered
? pg.reduce((a, osd, i) => a + (prev_hash[osd] == 1+i ? 1 : 0), 0)
: pg.reduce((a, osd) => a + (prev_hash[osd] ? 1 : 0), 0);
if (max_int < intersect_count)
{
max_int = intersect_count;
if (max_int >= pg_size)
{
break;
}
}
}
move_weights['pg_'+pg.join('_')] = pg_size-max_int;
}
}
}
return move_weights;
}
// Try to minimize data movement
async function optimize_change({ prev_pgs: prev_int_pgs, osd_weights, combinator, pg_size = 3, pg_minsize = 2, parity_space = 1, ordered = false })
{
if (!osd_weights)
{
return null;
}
// FIXME: use parity_chunks with parity_space instead of pg_minsize
let all_pgs = make_cyclic(combinator.random_combinations(), parity_space);
let pg_effsize = Object.values(all_pgs).reduce((a, c) => Math.max(a, c.filter(e => e != NO_OSD).length), 0);
pg_effsize = Math.min(pg_minsize, pg_effsize) + Math.max(0, Math.min(pg_size, pg_effsize) - pg_minsize) * parity_space;
const pg_count = prev_int_pgs.length;
const prev_weights = {};
const prev_pg_per_osd = {};
for (const pg of prev_int_pgs)
{
const pg_name = 'pg_'+pg.join('_');
prev_weights[pg_name] = (prev_weights[pg_name]||0) + 1;
for (let i = 0; i < pg.length; i++)
{
const osd = pg[i];
prev_pg_per_osd[osd] = prev_pg_per_osd[osd] || [];
prev_pg_per_osd[osd].push([ pg_name, (i >= pg_minsize ? parity_space : 1) ]);
}
}
const old_pg_size = prev_int_pgs[0].length;
// Get all combinations
if (old_pg_size == pg_size)
{
const still_valid = combinator.check_combinations(Object.keys(prev_weights).map(pg_name => pg_name.substr(3).split('_')));
for (const pg of still_valid)
{
all_pgs['pg_'+pg.join('_')] = pg;
}
}
all_pgs = Object.values(all_pgs);
const pg_per_osd = {};
for (const pg of all_pgs)
{
const pg_name = 'pg_'+pg.join('_');
for (let i = 0; i < pg.length; i++)
{
const osd = pg[i];
pg_per_osd[osd] = pg_per_osd[osd] || [];
pg_per_osd[osd].push([ pg_name, (i >= pg_minsize ? parity_space : 1) ]);
}
}
// Penalize PGs based on their similarity to old PGs
const move_weights = calc_intersect_weights(old_pg_size, pg_size, pg_count, prev_weights, all_pgs, ordered);
// Calculate total weight - old PG weights
const all_pg_names = all_pgs.map(pg => 'pg_'+pg.join('_'));
const all_pgs_hash = all_pg_names.reduce((a, c) => { a[c] = true; return a; }, {});
const total_weight = Object.values(osd_weights).reduce((a, c) => Number(a) + Number(c), 0);
// Generate the LP problem
let lp = '';
lp += 'max: '+all_pg_names.map(pg_name => (
prev_weights[pg_name] ? `${pg_size+1}*add_${pg_name} - ${pg_size+1}*del_${pg_name}` : `${pg_size+1-move_weights[pg_name]}*${pg_name}`
)).join(' + ')+';\n';
for (const osd in pg_per_osd)
{
if (osd !== NO_OSD)
{
const osd_sum = (pg_per_osd[osd]||[]).map(([ pg_name, space ]) => (
prev_weights[pg_name] ? `${space} * add_${pg_name} - ${space} * del_${pg_name}` : `${space} * ${pg_name}`
)).join(' + ');
const rm_osd_pg_count = (prev_pg_per_osd[osd]||[])
.reduce((a, [ old_pg_name, space ]) => (a + (all_pgs_hash[old_pg_name] ? space : 0)), 0);
const osd_pg_count = osd_weights[osd]*pg_effsize/total_weight*pg_count - rm_osd_pg_count;
lp += osd_sum + ' <= ' + osd_pg_count + ';\n';
}
}
let pg_vars = [];
for (const pg_name of all_pg_names)
{
if (prev_weights[pg_name])
{
pg_vars.push(`add_${pg_name}`, `del_${pg_name}`);
// Can't add or remove less than zero
lp += `add_${pg_name} >= 0;\n`;
lp += `del_${pg_name} >= 0;\n`;
// Can't remove more than the PG already has
lp += `add_${pg_name} - del_${pg_name} >= -${prev_weights[pg_name]};\n`;
}
else
{
pg_vars.push(pg_name);
lp += `${pg_name} >= 0;\n`;
}
}
lp += 'sec '+pg_vars.join(', ')+';\n';
// Solve it
const lp_result = await lp_solve(lp);
if (!lp_result)
{
console.log(lp);
throw new Error('Problem is infeasible or unbounded - is it a bug?');
}
// Generate the new distribution
const weights = { ...prev_weights };
for (const k in prev_weights)
{
if (!all_pgs_hash[k])
{
delete weights[k];
}
}
for (const k in lp_result.vars)
{
if (k.substr(0, 4) === 'add_')
{
weights[k.substr(4)] = (weights[k.substr(4)] || 0) + Number(lp_result.vars[k]);
}
else if (k.substr(0, 4) === 'del_')
{
weights[k.substr(4)] = (weights[k.substr(4)] || 0) - Number(lp_result.vars[k]);
}
else if (k.substr(0, 3) === 'pg_')
{
weights[k] = Number(lp_result.vars[k]);
}
}
for (const k in weights)
{
if (!weights[k])
{
delete weights[k];
}
}
const int_pgs = make_int_pgs(weights, pg_count);
// Align them with most similar previous PGs
const new_pgs = align_pgs(prev_int_pgs, int_pgs);
let differs = 0, osd_differs = 0;
for (let i = 0; i < pg_count; i++)
{
if (new_pgs[i].join('_') != prev_int_pgs[i].join('_'))
{
differs++;
}
}
if (ordered)
{
for (let i = 0; i < pg_count; i++)
{
for (let j = 0; j < pg_size; j++)
{
if (new_pgs[i][j] != prev_int_pgs[i][j])
{
osd_differs++;
}
}
}
}
else
{
for (let i = 0; i < pg_count; i++)
{
const old_map = prev_int_pgs[i].reduce((a, c) => { a[c] = (a[c]|0) + 1; return a; }, {});
for (let j = 0; j < pg_size; j++)
{
if ((0|old_map[new_pgs[i][j]]) > 0)
{
old_map[new_pgs[i][j]]--;
}
else
{
osd_differs++;
}
}
}
}
return {
prev_pgs: prev_int_pgs,
score: lp_result.score,
weights,
int_pgs: new_pgs,
differs,
osd_differs,
space: pg_effsize * pg_list_space_efficiency(new_pgs, osd_weights, pg_minsize, parity_space),
total_space: total_weight,
};
}
function print_change_stats(retval, detailed)
{
const new_pgs = retval.int_pgs;
const prev_int_pgs = retval.prev_pgs;
if (prev_int_pgs)
{
if (detailed)
{
for (let i = 0; i < new_pgs.length; i++)
{
if (new_pgs[i].join('_') != prev_int_pgs[i].join('_'))
{
console.log("pg "+i+": "+prev_int_pgs[i].join(' ')+" -> "+new_pgs[i].join(' '));
}
}
}
console.log(
"Data movement: "+retval.differs+" pgs, "+
retval.osd_differs+" pg*osds = "+Math.round(retval.osd_differs / prev_int_pgs.length / 3 * 10000)/100+" %"
);
}
console.log(
"Total space (raw): "+Math.round(retval.space*100)/100+" TB, space efficiency: "+
Math.round(retval.space/(retval.total_space||1)*10000)/100+" %"
);
}
function align_pgs(prev_int_pgs, int_pgs)
{
const aligned_pgs = [];
put_aligned_pgs(aligned_pgs, int_pgs, prev_int_pgs, (pg) => [ pg.join(':') ]);
put_aligned_pgs(aligned_pgs, int_pgs, prev_int_pgs, (pg) => [ pg[0]+'::'+pg[2], ':'+pg[1]+':'+pg[2], pg[0]+':'+pg[1]+':' ]);
put_aligned_pgs(aligned_pgs, int_pgs, prev_int_pgs, (pg) => [ pg[0]+'::', ':'+pg[1]+':', '::'+pg[2] ]);
const free_slots = prev_int_pgs.map((pg, i) => !aligned_pgs[i] ? i : null).filter(i => i != null);
for (const pg of int_pgs)
{
if (!free_slots.length)
{
throw new Error("Can't place unaligned PG");
}
aligned_pgs[free_slots.shift()] = pg;
}
return aligned_pgs;
}
function put_aligned_pgs(aligned_pgs, int_pgs, prev_int_pgs, keygen)
{
let prev_indexes = {};
for (let i = 0; i < prev_int_pgs.length; i++)
{
for (let k of keygen(prev_int_pgs[i]))
{
prev_indexes[k] = prev_indexes[k] || [];
prev_indexes[k].push(i);
}
}
PG: for (let i = int_pgs.length-1; i >= 0; i--)
{
let pg = int_pgs[i];
let keys = keygen(int_pgs[i]);
for (let k of keys)
{
while (prev_indexes[k] && prev_indexes[k].length)
{
let idx = prev_indexes[k].shift();
if (!aligned_pgs[idx])
{
aligned_pgs[idx] = pg;
int_pgs.splice(i, 1);
continue PG;
}
}
}
}
}
function pg_weights_space_efficiency(weights, pg_count, osd_sizes)
{
const per_osd = {};
for (const pg_name in weights)
{
for (const osd of pg_name.substr(3).split(/_/))
{
per_osd[osd] = (per_osd[osd]||0) + weights[pg_name];
}
}
return pg_per_osd_space_efficiency(per_osd, pg_count, osd_sizes);
}
function pg_list_space_efficiency(pgs, osd_sizes, pg_minsize, parity_space)
{
const per_osd = {};
for (const pg of pgs)
{
for (let i = 0; i < pg.length; i++)
{
const osd = pg[i];
per_osd[osd] = (per_osd[osd]||0) + (i >= pg_minsize ? (parity_space||1) : 1);
}
}
return pg_per_osd_space_efficiency(per_osd, pgs.length, osd_sizes);
}
function pg_per_osd_space_efficiency(per_osd, pg_count, osd_sizes)
{
// each PG gets randomly selected in 1/N cases
// & there are x PGs per OSD
// => an OSD is selected in x/N cases
// => total space * x/N <= OSD size
// => total space <= OSD size * N/x
let space;
for (let osd in per_osd)
{
if (osd in osd_sizes)
{
const space_estimate = osd_sizes[osd] * pg_count / per_osd[osd];
if (space == null || space > space_estimate)
{
space = space_estimate;
}
}
}
return space == null ? 0 : space;
}
module.exports = {
NO_OSD,
optimize_initial,
optimize_change,
print_change_stats,
pg_weights_space_efficiency,
pg_list_space_efficiency,
pg_per_osd_space_efficiency,
lp_solve,
make_int_pgs,
align_pgs,
};
+38
View File
@@ -0,0 +1,38 @@
function select_murmur3(count, cb)
{
if (!count)
{
return null;
}
else
{
let i = 0, maxh = -1;
for (let j = 0; j < count; j++)
{
const h = murmur3(cb(j));
if (h > maxh)
{
i = j;
maxh = h;
}
}
return i;
}
}
function murmur3(s)
{
let hash = 0x12345678;
for (let i = 0; i < s.length; i++)
{
hash ^= s.charCodeAt(i);
hash = (hash*0x5bd1e995) & 0xFFFFFFFF;
hash ^= (hash >> 15);
}
return hash;
}
module.exports = {
murmur3,
select_murmur3,
};
+241
View File
@@ -0,0 +1,241 @@
const { select_murmur3 } = require('./murmur3.js');
const NO_OSD = 'Z';
class SimpleCombinator
{
constructor(flat_tree, pg_size, max_combinations, ordered)
{
this.osd_tree = flat_tree;
this.pg_size = pg_size;
this.max_combinations = max_combinations;
this.ordered = ordered;
}
random_combinations()
{
return random_combinations(this.osd_tree, this.pg_size, this.max_combinations, this.ordered);
}
check_combinations(pgs)
{
return check_combinations(this.osd_tree, pgs);
}
}
// Convert multi-level osd_tree = { level: number|string, id?: string, size?: number, children?: osd_tree }[]
// levels = { string: number }
// to a two-level osd_tree suitable for all_combinations()
function flatten_tree(osd_tree, levels, failure_domain_level, osd_level, domains = {}, i = { i: 1 })
{
osd_level = levels[osd_level] || osd_level;
failure_domain_level = levels[failure_domain_level] || failure_domain_level;
for (const node of osd_tree)
{
if ((levels[node.level] || node.level) < failure_domain_level)
{
flatten_tree(node.children||[], levels, failure_domain_level, osd_level, domains, i);
}
else
{
domains['dom'+(i.i++)] = extract_osds([ node ], levels, osd_level);
}
}
return domains;
}
function extract_osds(osd_tree, levels, osd_level, osds = {})
{
for (const node of osd_tree)
{
if ((levels[node.level] || node.level) >= osd_level)
{
osds[node.id] = node.size;
}
else
{
extract_osds(node.children||[], levels, osd_level, osds);
}
}
return osds;
}
// ordered = don't treat (x,y) and (y,x) as equal
function random_combinations(osd_tree, pg_size, count, ordered)
{
const osds = Object.keys(osd_tree).reduce((a, c) => { a[c] = Object.keys(osd_tree[c]).sort(); return a; }, {});
const hosts = Object.keys(osd_tree).sort().filter(h => osds[h].length > 0);
const r = {};
// Generate random combinations including each OSD at least once
for (let h = 0; h < hosts.length; h++)
{
for (let o = 0; o < osds[hosts[h]].length; o++)
{
const pg = [ osds[hosts[h]][o] ];
const cur_hosts = [ ...hosts ];
cur_hosts.splice(h, 1);
for (let i = 1; i < pg_size && i < hosts.length; i++)
{
const next_host = select_murmur3(cur_hosts.length, i => pg[0]+':i:'+cur_hosts[i]);
const next_osd = select_murmur3(osds[cur_hosts[next_host]].length, i => pg[0]+':i:'+osds[cur_hosts[next_host]][i]);
pg.push(osds[cur_hosts[next_host]][next_osd]);
cur_hosts.splice(next_host, 1);
}
while (pg.length < pg_size)
{
pg.push(NO_OSD);
}
r['pg_'+pg.join('_')] = pg;
}
}
// Generate purely random combinations
while (count > 0)
{
let host_idx = [];
const cur_hosts = [ ...hosts.map((h, i) => i) ];
const max_hosts = pg_size < hosts.length ? pg_size : hosts.length;
if (ordered)
{
for (let i = 0; i < max_hosts; i++)
{
const r = select_murmur3(cur_hosts.length, i => count+':h:'+cur_hosts[i]);
host_idx[i] = cur_hosts[r];
cur_hosts.splice(r, 1);
}
}
else
{
for (let i = 0; i < max_hosts; i++)
{
const r = select_murmur3(cur_hosts.length - (max_hosts - i - 1), i => count+':h:'+cur_hosts[i]);
host_idx[i] = cur_hosts[r];
cur_hosts.splice(0, r+1);
}
}
let pg = host_idx.map(h => osds[hosts[h]][select_murmur3(osds[hosts[h]].length, i => count+':o:'+osds[hosts[h]][i])]);
while (pg.length < pg_size)
{
pg.push(NO_OSD);
}
r['pg_'+pg.join('_')] = pg;
count--;
}
return r;
}
// Super-stupid algorithm. Given the current OSD tree, generate all possible OSD combinations
// osd_tree = { failure_domain1: { osd1: size1, ... }, ... }
// ordered = return combinations without duplicates having different order
function all_combinations(osd_tree, pg_size, ordered, count)
{
const hosts = Object.keys(osd_tree).sort();
const osds = Object.keys(osd_tree).reduce((a, c) => { a[c] = Object.keys(osd_tree[c]).sort(); return a; }, {});
while (hosts.length < pg_size)
{
osds[NO_OSD] = [ NO_OSD ];
hosts.push(NO_OSD);
}
let host_idx = [];
let osd_idx = [];
for (let i = 0; i < pg_size; i++)
{
host_idx.push(i);
osd_idx.push(0);
}
const r = [];
while (!count || count < 0 || r.length < count)
{
r.push(host_idx.map((hi, i) => osds[hosts[hi]][osd_idx[i]]));
let inc = pg_size-1;
while (inc >= 0)
{
osd_idx[inc]++;
if (osd_idx[inc] >= osds[hosts[host_idx[inc]]].length)
{
osd_idx[inc] = 0;
inc--;
}
else
{
break;
}
}
if (inc < 0)
{
// no osds left in the current host combination, select the next one
inc = pg_size-1;
same_again: while (inc >= 0)
{
host_idx[inc]++;
for (let prev_host = 0; prev_host < inc; prev_host++)
{
if (host_idx[prev_host] == host_idx[inc])
{
continue same_again;
}
}
if (host_idx[inc] < (ordered ? hosts.length-(pg_size-1-inc) : hosts.length))
{
while ((++inc) < pg_size)
{
host_idx[inc] = (ordered ? host_idx[inc-1]+1 : 0);
}
break;
}
else
{
inc--;
}
}
if (inc < 0)
{
break;
}
}
}
return r;
}
function check_combinations(osd_tree, pgs)
{
const host_per_osd = {};
for (const host in osd_tree)
{
for (const osd in osd_tree[host])
{
host_per_osd[osd] = host;
}
}
const res = [];
skip_pg: for (const pg of pgs)
{
const seen_hosts = {};
for (const osd of pg)
{
if (!host_per_osd[osd] || seen_hosts[host_per_osd[osd]])
{
continue skip_pg;
}
seen_hosts[host_per_osd[osd]] = true;
}
res.push(pg);
}
return res;
}
function compat(params)
{
return {
...params,
osd_weights: Object.assign({}, ...Object.values(params.osd_tree)),
combinator: new SimpleCombinator(params.osd_tree, params.pg_size, params.max_combinations||10000),
};
}
module.exports = {
flatten_tree,
all_combinations,
SimpleCombinator,
compat,
NO_OSD,
};
+132
View File
@@ -0,0 +1,132 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
// Interesting real-world example coming from Ceph with EC and compression enabled.
// EC parity chunks can't be compressed as efficiently as data chunks,
// thus they occupy more space (2.26x more space) in OSD object stores.
// This leads to really uneven OSD fill ratio in Ceph even when PGs are perfectly balanced.
// But we support this case with the "parity_space" parameter in optimize_initial()/optimize_change().
const { SimpleCombinator } = require('./simple_pgs.js');
const LPOptimizer = require('./lp_optimizer.js');
const osd_tree = {
ripper5: {
osd0: 3.493144989013672,
osd1: 3.493144989013672,
osd2: 3.454082489013672,
osd12: 3.461894989013672,
},
ripper7: {
osd4: 3.638690948486328,
osd5: 3.638690948486328,
osd6: 3.638690948486328,
},
ripper4: {
osd9: 3.4609375,
osd10: 3.4609375,
osd11: 3.4609375,
},
ripper6: {
osd3: 3.5849609375,
osd7: 3.5859336853027344,
osd8: 3.638690948486328,
osd13: 3.461894989013672
},
};
const prev_pgs = [[12,7,5],[6,11,12],[3,6,9],[10,0,5],[2,5,13],[9,8,6],[3,4,12],[7,4,12],[12,11,13],[13,6,0],[4,13,10],[9,7,6],[7,10,0],[10,8,0],[3,10,2],[3,0,4],[6,13,0],[13,10,0],[13,10,5],[8,11,6],[3,9,2],[2,8,5],[8,9,5],[3,12,11],[0,7,4],[13,11,1],[11,3,12],[12,8,10],[7,5,12],[2,13,5],[7,11,0],[13,2,6],[0,6,8],[13,1,6],[0,13,4],[0,8,10],[4,10,0],[8,12,4],[8,12,9],[12,7,4],[13,9,5],[3,2,11],[1,9,7],[1,8,5],[5,12,9],[3,5,12],[2,8,10],[0,8,4],[1,4,11],[7,10,2],[12,13,5],[3,1,11],[7,1,4],[4,12,8],[7,0,9],[11,1,8],[3,0,5],[11,13,0],[1,13,5],[12,7,10],[12,8,4],[11,13,5],[0,11,6],[2,11,3],[13,1,11],[2,7,10],[7,10,12],[7,12,10],[12,11,5],[13,12,10],[2,3,9],[4,3,9],[13,2,5],[7,12,6],[12,10,13],[9,8,1],[13,1,5],[9,5,12],[5,11,7],[6,2,9],[8,11,6],[12,5,8],[6,13,1],[7,6,11],[2,3,6],[8,5,9],[1,13,6],[9,3,2],[7,11,1],[3,10,1],[0,11,7],[3,0,5],[1,3,6],[6,0,9],[3,11,4],[8,10,2],[13,1,9],[12,6,9],[3,12,9],[12,8,9],[7,5,0],[8,12,5],[0,11,3],[12,11,13],[0,7,11],[0,3,10],[1,3,11],[2,7,11],[13,2,6],[9,12,13],[8,2,4],[0,7,4],[5,13,0],[13,12,9],[1,9,8],[0,10,3],[3,5,10],[7,12,9],[2,13,4],[12,7,5],[9,2,7],[3,2,9],[6,2,7],[3,1,9],[4,3,2],[5,3,11],[0,7,6],[1,6,13],[7,10,2],[12,4,8],[13,12,6],[7,5,11],[6,2,3],[2,7,6],[2,3,10],[2,7,10],[11,12,6],[0,13,5],[10,2,4],[13,0,11],[7,0,6],[8,9,4],[8,4,11],[7,11,2],[3,4,2],[6,1,3],[7,2,11],[8,9,4],[11,4,8],[10,3,1],[2,10,13],[1,7,11],[13,11,12],[2,6,9],[10,0,13],[7,10,4],[0,11,13],[13,10,1],[7,5,0],[7,12,10],[3,1,4],[7,1,5],[3,11,5],[7,5,0],[1,3,5],[10,5,12],[0,3,9],[7,1,11],[11,8,12],[3,6,2],[7,12,9],[7,11,12],[4,11,3],[0,11,13],[13,2,5],[1,5,8],[0,11,8],[3,5,1],[11,0,6],[3,11,2],[11,8,12],[4,1,3],[10,13,4],[13,9,6],[2,3,10],[12,7,9],[10,0,4],[10,13,2],[3,11,1],[7,2,9],[1,7,4],[13,1,4],[7,0,6],[5,3,9],[10,0,7],[0,7,10],[3,6,10],[13,0,5],[8,4,1],[3,1,10],[2,10,13],[13,0,5],[13,10,2],[12,7,9],[6,8,10],[6,1,8],[10,8,1],[13,5,0],[5,11,3],[7,6,1],[8,5,9],[2,13,11],[10,12,4],[13,4,1],[2,13,4],[11,7,0],[2,9,7],[1,7,6],[8,0,4],[8,1,9],[7,10,12],[13,9,6],[7,6,11],[13,0,4],[1,8,4],[3,12,5],[10,3,1],[10,2,13],[2,4,8],[6,2,3],[3,0,10],[6,7,12],[8,12,5],[3,0,6],[13,12,10],[11,3,6],[9,0,13],[10,0,6],[7,5,2],[1,3,11],[7,10,2],[2,9,8],[11,13,12],[0,8,4],[8,12,11],[6,0,3],[1,13,4],[11,8,2],[12,3,6],[4,7,1],[7,6,12],[3,10,6],[0,10,7],[8,9,1],[0,10,6],[8,10,1]]
.map(pg => pg.map(n => 'osd'+n));
const by_osd = {};
for (let i = 0; i < prev_pgs.length; i++)
{
for (let j = 0; j < prev_pgs[i].length; j++)
{
by_osd[prev_pgs[i][j]] = by_osd[prev_pgs[i][j]] || [];
by_osd[prev_pgs[i][j]][j] = (by_osd[prev_pgs[i][j]][j] || 0) + 1;
}
}
/*
This set of PGs was balanced by hand, by heavily tuning OSD weights in Ceph:
{
osd0: 4.2,
osd1: 3.5,
osd2: 3.45409,
osd3: 4.5,
osd4: 1.4,
osd5: 1.4,
osd6: 1.75,
osd7: 4.5,
osd8: 4.4,
osd9: 2.2,
osd10: 2.7,
osd11: 2,
osd12: 3.4,
osd13: 3.4,
}
EC+compression is a nightmare in Ceph, yeah :))
To calculate the average ratio between data chunks and parity chunks we
calculate the number of PG chunks for each chunk role for each OSD:
{
osd12: [ 18, 22, 17 ],
osd7: [ 35, 22, 8 ],
osd5: [ 6, 17, 27 ],
osd6: [ 13, 12, 28 ],
osd11: [ 13, 26, 20 ],
osd3: [ 30, 20, 10 ],
osd9: [ 8, 12, 26 ],
osd10: [ 15, 23, 20 ],
osd0: [ 22, 22, 14 ],
osd2: [ 22, 16, 16 ],
osd13: [ 29, 19, 13 ],
osd8: [ 20, 18, 12 ],
osd4: [ 8, 10, 28 ],
osd1: [ 17, 17, 17 ]
}
And now we can pick a pair of OSDs and determine the ratio by solving the following:
osd5 = 23*X + 27*Y = 3249728140
osd13 = 48*X + 13*Y = 2991675992
=>
osd5 - 27/13*osd13 = 23*X - 27/13*48*X = -76.6923076923077*X = -2963752766.46154
=>
X = 38644720.1243731
Y = (osd5-23*X)/27 = 87440725.0792377
Y/X = 2.26268232239284 ~= 2.26
Which means that parity chunks are compressed ~2.26 times worse than data chunks.
Fine, let's try to optimize for it.
*/
async function run()
{
const osd_weights = Object.assign({}, ...Object.values(osd_tree));
const total_weight = Object.values(osd_weights).reduce((a, c) => Number(a) + Number(c), 0);
const eff = LPOptimizer.pg_list_space_efficiency(prev_pgs, osd_weights, 2, 2.26);
const orig = eff*4.26 / total_weight;
console.log('Original efficiency was: '+Math.round(orig*10000)/100+' %');
const combinator = new SimpleCombinator(osd_tree, 3, 10000);
let prev = await LPOptimizer.optimize_initial({ osd_weights, combinator, pg_size: 3, pg_count: 256, parity_space: 2.26 });
LPOptimizer.print_change_stats(prev);
let next = await LPOptimizer.optimize_change({ prev_pgs, osd_weights, combinator, pg_size: 3, parity_space: 2.26 });
LPOptimizer.print_change_stats(next);
}
run().catch(console.error);
+50
View File
@@ -0,0 +1,50 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
const { compat } = require('./simple_pgs.js');
const LPOptimizer = require('./lp_optimizer.js');
async function run()
{
const osd_tree = {
100: { 1: 1 },
200: { 2: 1 },
300: { 3: 1 },
};
let res;
console.log('16 PGs, size=3');
res = await LPOptimizer.optimize_initial(compat({ osd_tree, pg_size: 3, pg_count: 16, ordered: false }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 3, 'Initial distribution');
console.log('\nChange size to 2');
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2, ordered: false }));
LPOptimizer.print_change_stats(res, false);
assert(res.space >= 3*14/16 && res.osd_differs == 0, 'Redistribution');
console.log('\nRemove OSD 3');
const no3_tree = { ...osd_tree };
delete no3_tree['300'];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree: no3_tree, pg_size: 2, ordered: false }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 2, 'Redistribution after OSD removal');
console.log('\n16 PGs, size=3, ordered');
res = await LPOptimizer.optimize_initial(compat({ osd_tree, pg_size: 3, pg_count: 16, ordered: true }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 3, 'Initial distribution');
console.log('\nChange size to 2, ordered');
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2, ordered: true }));
LPOptimizer.print_change_stats(res, false);
assert(res.space >= 3*14/16 && res.osd_differs < 8, 'Redistribution');
}
function assert(cond, txt)
{
if (!cond)
{
throw new Error((txt||'test')+' failed');
}
}
run().catch(console.error);
@@ -0,0 +1,90 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
const { compat, flatten_tree } = require('./simple_pgs.js');
const LPOptimizer = require('./lp_optimizer.js');
const crush_tree = [
{ level: 1, children: [
{ level: 2, children: [
{ level: 3, id: 1, size: 3 },
{ level: 3, id: 2, size: 3 },
] },
{ level: 2, children: [
{ level: 3, id: 3, size: 3 },
{ level: 3, id: 4, size: 3 },
] },
] },
{ level: 1, children: [
{ level: 2, children: [
{ level: 3, id: 5, size: 3 },
{ level: 3, id: 6, size: 3 },
] },
{ level: 2, children: [
{ level: 3, id: 7, size: 3 },
{ level: 3, id: 8, size: 3 },
] },
] },
{ level: 1, children: [
{ level: 2, children: [
{ level: 3, id: 9, size: 3 },
{ level: 3, id: 10, size: 3 },
] },
{ level: 2, children: [
{ level: 3, id: 11, size: 3 },
{ level: 3, id: 12, size: 3 },
] },
] },
];
const osd_tree = flatten_tree(crush_tree, {}, 1, 3);
console.log(osd_tree);
async function run()
{
const cur_tree = {};
console.log('Empty tree:');
let res = await LPOptimizer.optimize_initial(compat({ osd_tree: cur_tree, pg_size: 3, pg_count: 256 }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 0);
console.log('\nAdding 1st failure domain:');
cur_tree['dom1'] = osd_tree['dom1'];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 12 && res.total_space == 12);
console.log('\nAdding 2nd failure domain:');
cur_tree['dom2'] = osd_tree['dom2'];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 24 && res.total_space == 24);
console.log('\nAdding 3rd failure domain:');
cur_tree['dom3'] = osd_tree['dom3'];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 36 && res.total_space == 36);
console.log('\nRemoving 3rd failure domain:');
delete cur_tree['dom3'];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 24 && res.total_space == 24);
console.log('\nRemoving 2nd failure domain:');
delete cur_tree['dom2'];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 12 && res.total_space == 12);
console.log('\nRemoving 1st failure domain:');
delete cur_tree['dom1'];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 }));
LPOptimizer.print_change_stats(res, false);
assert(res.space == 0);
}
function assert(cond, txt)
{
if (!cond)
{
throw new Error((txt||'test')+' failed');
}
}
run().catch(console.error);
@@ -0,0 +1,34 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
const { compat } = require('./simple_pgs.js');
const LPOptimizer = require('./lp_optimizer.js');
const osd_tree = {
100: {
1: 0.1,
2: 0.1,
3: 0.1,
},
200: {
4: 0.1,
5: 0.1,
6: 0.1,
},
};
async function run()
{
let res;
console.log('256 PGs, 3+3 OSDs, size=2');
res = await LPOptimizer.optimize_initial(compat({ osd_tree, pg_size: 2, pg_count: 256 }));
LPOptimizer.print_change_stats(res, false);
// Should NOT fail with the "unfeasible or unbounded" exception
console.log('\nRemoving osd.2');
delete osd_tree[100][2];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 }));
LPOptimizer.print_change_stats(res, false);
}
run().catch(console.error);
+116
View File
@@ -0,0 +1,116 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
const { compat, flatten_tree } = require('./simple_pgs.js');
const LPOptimizer = require('./lp_optimizer.js');
const osd_tree = {
100: {
7: 3.63869,
},
300: {
10: 3.46089,
11: 3.46089,
12: 3.46089,
},
400: {
1: 3.49309,
2: 3.49309,
3: 3.49309,
},
500: {
4: 3.58498,
/*8: 3.58589,*/
9: 3.63869,
},
600: {
5: 3.63869,
6: 3.63869,
},
/* 100: {
1: 2.72800,
},
200: {
2: 2.72900,
},
300: {
3: 1.87000,
},
400: {
4: 1.87000,
},
500: {
5: 3.63869,
},*/
};
const crush_tree = [
{ level: 1, children: [
{ level: 2, children: [
{ level: 3, id: 1, size: 3 },
{ level: 3, id: 2, size: 2 },
] },
{ level: 2, children: [
{ level: 3, id: 3, size: 4 },
{ level: 3, id: 4, size: 4 },
] },
] },
{ level: 1, children: [
{ level: 2, children: [
{ level: 3, id: 5, size: 4 },
{ level: 3, id: 6, size: 1 },
] },
{ level: 2, children: [
{ level: 3, id: 7, size: 3 },
{ level: 3, id: 8, size: 5 },
] },
] },
{ level: 1, children: [
{ level: 2, children: [
{ level: 3, id: 9, size: 5 },
{ level: 3, id: 10, size: 2 },
] },
{ level: 2, children: [
{ level: 3, id: 11, size: 3 },
{ level: 3, id: 12, size: 3 },
] },
] },
];
async function run()
{
let res;
// Test: add 1 OSD of almost the same size. Ideal data movement could be 1/12 = 8.33%. Actual is ~13%
// Space efficiency is ~99% in all cases.
console.log('256 PGs, size=2');
res = await LPOptimizer.optimize_initial(compat({ osd_tree, pg_size: 2, pg_count: 256 }));
LPOptimizer.print_change_stats(res, false);
console.log('\nAdding osd.8');
osd_tree[500][8] = 3.58589;
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 }));
LPOptimizer.print_change_stats(res, false);
console.log('\nRemoving osd.8');
delete osd_tree[500][8];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 }));
LPOptimizer.print_change_stats(res, false);
console.log('\n256 PGs, size=3');
res = await LPOptimizer.optimize_initial(compat({ osd_tree, pg_size: 3, pg_count: 256 }));
LPOptimizer.print_change_stats(res, false);
console.log('\nAdding osd.8');
osd_tree[500][8] = 3.58589;
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree, pg_size: 3 }));
LPOptimizer.print_change_stats(res, false);
console.log('\nRemoving osd.8');
delete osd_tree[500][8];
res = await LPOptimizer.optimize_change(compat({ prev_pgs: res.int_pgs, osd_tree, pg_size: 3 }));
LPOptimizer.print_change_stats(res, false);
console.log('\n256 PGs, size=3, failure domain=rack');
res = await LPOptimizer.optimize_initial(compat({ osd_tree: flatten_tree(crush_tree, {}, 1, 3), pg_size: 3, pg_count: 256 }));
LPOptimizer.print_change_stats(res, false);
}
run().catch(console.error);
+118
View File
@@ -0,0 +1,118 @@
const { random_custom_combinations, index_tree, parse_level_indexes, parse_pg_dsl } = require('./dsl_pgs.js');
function check(result, expected)
{
console.dir(result, { depth: null });
if (JSON.stringify(result) !== JSON.stringify(expected))
{
process.stderr.write('Unexpected value, expected: ');
console.dir(expected, { depth: null });
process.exit(1);
}
}
check(
parse_pg_dsl("any, dc=1 host!=1, dc!=1, dc=3 host!=3, dc!=(1,3), dc=5 host!=5"),
[
[],
[ [ 'dc', '=', 1 ], [ 'host', '!=', 1 ] ],
[ [ 'dc', '!=', 1 ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', 3 ] ],
[ [ 'dc', '!=', [ 1, 3 ] ] ],
[ [ 'dc', '=', 5 ], [ 'host', '!=', 5 ] ],
]
);
check(
parse_pg_dsl("dc=meow, dc!=1, dc>2"),
[
[ [ 'dc', '=', { id: 'meow' } ] ],
[ [ 'dc', '!=', 1 ] ],
[ [ 'dc', '>', 2 ] ],
]
);
check(
parse_level_indexes({ dc: '112233', host: 'ABCDEF' }),
[
[],
[ [ 'dc', '=', 1 ], [ 'host', '!=', [ 1 ] ] ],
[ [ 'dc', '!=', [ 1 ] ], [ 'host', '!=', [ 1, 2 ] ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', [ 1, 2, 3 ] ] ],
[ [ 'dc', '!=', [ 1, 3 ] ], [ 'host', '!=', [ 1, 2, 3, 4 ] ] ],
[ [ 'dc', '=', 5 ], [ 'host', '!=', [ 1, 2, 3, 4, 5 ] ] ],
]
);
check(
parse_level_indexes({ dc: '112233', host: 'ABCDEF' }, [ 'dc', 'host' ]),
[
[],
[ [ 'dc', '=', 1 ], [ 'host', '!=', [ 1 ] ] ],
[ [ 'dc', '!=', [ 1 ] ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', [ 3 ] ] ],
[ [ 'dc', '!=', [ 1, 3 ] ] ],
[ [ 'dc', '=', 5 ], [ 'host', '!=', [ 5 ] ] ],
]
);
check(
parse_level_indexes({ dc: '112211223333', host: '123456789ABC' }),
[
[],
[ [ 'dc', '=', 1 ], [ 'host', '!=', [ 1 ] ] ],
[ [ 'dc', '!=', [ 1 ] ], [ 'host', '!=', [ 1, 2 ] ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', [ 1, 2, 3 ] ] ],
[ [ 'dc', '=', 1 ], [ 'host', '!=', [ 1, 2, 3, 4 ] ] ],
[ [ 'dc', '=', 1 ], [ 'host', '!=', [ 1, 2, 3, 4, 5 ] ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', [ 1, 2, 3, 4, 5, 6 ] ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', [ 1, 2, 3, 4, 5, 6, 7 ] ] ],
[ [ 'dc', '!=', [ 1, 3 ] ], [ 'host', '!=', [ 1, 2, 3, 4, 5, 6, 7, 8 ] ] ],
[ [ 'dc', '=', 9 ], [ 'host', '!=', [ 1, 2, 3, 4, 5, 6, 7, 8, 9 ] ] ],
[ [ 'dc', '=', 9 ], [ 'host', '!=', [ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 ] ] ],
[ [ 'dc', '=', 9 ], [ 'host', '!=', [ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 ] ] ],
]
);
check(
parse_level_indexes({ dc: '112211223333', host: '123456789ABC' }, [ 'dc', 'host' ]),
[
[],
[ [ 'dc', '=', 1 ], [ 'host', '!=', [ 1 ] ] ],
[ [ 'dc', '!=', [ 1 ] ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', [ 3 ] ] ],
[ [ 'dc', '=', 1 ], [ 'host', '!=', [ 1, 2 ] ] ],
[ [ 'dc', '=', 1 ], [ 'host', '!=', [ 1, 2, 5 ] ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', [ 3, 4 ] ] ],
[ [ 'dc', '=', 3 ], [ 'host', '!=', [ 3, 4, 7 ] ] ],
[ [ 'dc', '!=', [ 1, 3 ] ] ],
[ [ 'dc', '=', 9 ], [ 'host', '!=', [ 9 ] ] ],
[ [ 'dc', '=', 9 ], [ 'host', '!=', [ 9, 10 ] ] ],
[ [ 'dc', '=', 9 ], [ 'host', '!=', [ 9, 10, 11 ] ] ]
]
);
check(
Object.keys(random_custom_combinations(index_tree([
{ id: '1', size: 1, level: 'osd' },
{ id: '2', size: 2, level: 'osd' },
{ id: '3', size: 3, level: 'osd' }
]), parse_level_indexes({ osd: '12' }), 10000)).sort(),
[ 'pg_1_2', 'pg_1_3', 'pg_2_3' ]
);
check(
Object.keys(random_custom_combinations(index_tree([
{ id: 'h1', level: 'host' },
{ id: 'h2', level: 'host' },
{ id: 'h3', level: 'host' },
{ id: '1', size: 1, level: 'osd', parent: 'h1' },
{ id: '2', size: 1, level: 'osd', parent: 'h2' },
{ id: '3', size: 1, level: 'osd', parent: 'h2' },
{ id: '4', size: 1, level: 'osd', parent: 'h3' },
{ id: '5', size: 1, level: 'osd', parent: 'h3' },
]), parse_level_indexes({ host: '1122', osd: '1234' }), 10000)).sort(),
[ 'pg_2_3_4_5' ]
);
console.log('OK');