public static async Task Sampling(string[] path, IVarTuple extras, IFdbDatabase db, TextWriter log, CancellationToken ct)
{
double ratio = 0.1d;
bool auto = true;
if (extras.Count > 0)
{
double x = extras.Get <double>(0);
if (x > 0 && x <= 1)
{
ratio = x;
}
auto = false;
}
var folder = await TryOpenCurrentDirectoryAsync(path, db, ct);
KeyRange span;
if (folder is FdbDirectorySubspace)
{
span = KeyRange.StartsWith((folder as FdbDirectorySubspace).Copy().GetPrefix());
log.WriteLine($"Reading list of shards for /{String.Join("/", path)} under {FdbKey.Dump(span.Begin)} ...");
}
else
{
log.WriteLine("Reading list of shards for the whole cluster ...");
span = KeyRange.All;
}
// dump keyServers
var ranges = await Fdb.System.GetChunksAsync(db, span, ct);
log.WriteLine($"> Found {ranges.Count:N0} shard(s)");
// take a sample
var samples = new List <KeyRange>();
if (ranges.Count <= 32)
{ // small enough to scan it all
samples.AddRange(ranges);
log.WriteLine($"Sampling all {samples.Count:N0} shards ...");
}
else
{ // need to take a random subset
var rnd = new Random();
int sz = Math.Max((int)Math.Ceiling(ratio * ranges.Count), 1);
if (auto)
{
if (sz > 100)
{
sz = 100; //SAFETY
}
if (sz < 32)
{
sz = Math.Max(sz, Math.Min(32, ranges.Count));
}
}
var population = new List <KeyRange>(ranges);
for (int i = 0; i < sz; i++)
{
int p = rnd.Next(population.Count);
samples.Add(population[p]);
population.RemoveAt(p);
}
log.WriteLine($"Sampling {samples.Count:N0} out of {ranges.Count:N0} shards ({(100.0 * samples.Count / ranges.Count):N1}%) ...");
}
log.WriteLine();
const string FORMAT_STRING = "{0,9} ║{1,10}{6,6} {2,-29} ║{3,10}{7,7} {4,-37} ║{5,10}";
const string SCALE_KEY = "....--------========########M";
const string SCALE_VAL = "....--------========########@@@@@@@@M";
log.WriteLine(FORMAT_STRING, "Count", "Keys", SCALE_KEY, "Values", SCALE_VAL, "Total", "med.", "med.");
var rangeOptions = new FdbRangeOptions {
Mode = FdbStreamingMode.WantAll
};
samples = samples.OrderBy(x => x.Begin).ToList();
long globalSize = 0;
long globalCount = 0;
int workers = 8; // Math.Max(4, Environment.ProcessorCount);
var sw = Stopwatch.StartNew();
var tasks = new List <Task>();
int n = samples.Count;
while (samples.Count > 0)
{
while (tasks.Count < workers && samples.Count > 0)
{
var range = samples[0];
samples.RemoveAt(0);
tasks.Add(Task.Run(async() =>
{
var kk = new RobustHistogram(RobustHistogram.TimeScale.Ticks);
var vv = new RobustHistogram(RobustHistogram.TimeScale.Ticks);
#region Method 1: get_range everything...
using (var tr = db.BeginTransaction(ct))
{
long keySize = 0;
long valueSize = 0;
long count = 0;
int iter = 0;
var beginSelector = KeySelector.FirstGreaterOrEqual(range.Begin);
var endSelector = KeySelector.FirstGreaterOrEqual(range.End);
while (true)
{
FdbRangeChunk data = default(FdbRangeChunk);
FdbException error = null;
try
{
data = await tr.Snapshot.GetRangeAsync(
beginSelector,
endSelector,
rangeOptions,
iter
).ConfigureAwait(false);
}
catch (FdbException e)
{
error = e;
}
if (error != null)
{
await tr.OnErrorAsync(error.Code).ConfigureAwait(false);
continue;
}
if (data.Count == 0)
{
break;
}
count += data.Count;
foreach (var kvp in data)
{
keySize += kvp.Key.Count;
valueSize += kvp.Value.Count;
kk.Add(TimeSpan.FromTicks(kvp.Key.Count));
vv.Add(TimeSpan.FromTicks(kvp.Value.Count));
}
if (!data.HasMore)
{
break;
}
beginSelector = KeySelector.FirstGreaterThan(data.Last);
++iter;
}
long totalSize = keySize + valueSize;
Interlocked.Add(ref globalSize, totalSize);
Interlocked.Add(ref globalCount, count);
lock (log)
{
log.WriteLine(FORMAT_STRING, count.ToString("N0"), FormatSize(keySize), kk.GetDistribution(begin: 1, end: 12000, fold: 2), FormatSize(valueSize), vv.GetDistribution(begin: 1, end: 120000, fold: 2), FormatSize(totalSize), FormatSize((int)Math.Ceiling(kk.Median)), FormatSize((int)Math.Ceiling(vv.Median)));
}
}
#endregion
#region Method 2: estimate the count using key selectors...
//long counter = await Fdb.System.EstimateCountAsync(db, range, ct);
//Console.WriteLine("COUNT = " + counter.ToString("N0"));
#endregion
}, ct));
}
var done = await Task.WhenAny(tasks);
tasks.Remove(done);
}
await Task.WhenAll(tasks);
sw.Stop();
log.WriteLine();
if (n != ranges.Count)
{
log.WriteLine($"Sampled {FormatSize(globalSize)} ({globalSize:N0} bytes) and {globalCount:N0} keys in {sw.Elapsed.TotalSeconds:N1} sec");
log.WriteLine($"> Estimated total size is {FormatSize(globalSize * ranges.Count / n)}");
}
else
{
log.WriteLine($"Found {FormatSize(globalSize)} ({globalSize:N0} bytes) and {globalCount:N0} keys in {sw.Elapsed.TotalSeconds:N1} sec");
// compare to the whole cluster
ranges = await Fdb.System.GetChunksAsync(db, FdbKey.MinValue, FdbKey.MaxValue, ct);
log.WriteLine($"> This directory contains ~{(100.0 * n / ranges.Count):N2}% of all data");
}
log.WriteLine();
}
public static async Task Sampling(string[] path, IFdbTuple extras, IFdbDatabase db, TextWriter log, CancellationToken ct)
{
double ratio = 0.1d;
bool auto = true;
if (extras.Count > 0)
{
double x = extras.Get<double>(0);
if (x > 0 && x <= 1) ratio = x;
auto = false;
}
var folder = await TryOpenCurrentDirectoryAsync(path, db, ct);
FdbKeyRange span;
if (folder is FdbDirectorySubspace)
{
span = FdbKeyRange.StartsWith((folder as FdbDirectorySubspace).Copy());
log.WriteLine("Reading list of shards for /{0} under {1} ...", String.Join("/", path), FdbKey.Dump(span.Begin));
}
else
{
log.WriteLine("Reading list of shards for the whole cluster ...");
span = FdbKeyRange.All;
}
// dump keyServers
var ranges = await Fdb.System.GetChunksAsync(db, span, ct);
log.WriteLine("> Found {0:N0} shard(s)", ranges.Count);
// take a sample
var samples = new List<FdbKeyRange>();
if (ranges.Count <= 32)
{ // small enough to scan it all
samples.AddRange(ranges);
log.WriteLine("Sampling all {0:N0} shards ...", samples.Count);
}
else
{ // need to take a random subset
var rnd = new Random();
int sz = Math.Max((int)Math.Ceiling(ratio * ranges.Count), 1);
if (auto)
{
if (sz > 100) sz = 100; //SAFETY
if (sz < 32) sz = Math.Max(sz, Math.Min(32, ranges.Count));
}
var population = new List<FdbKeyRange>(ranges);
for (int i = 0; i < sz; i++)
{
int p = rnd.Next(population.Count);
samples.Add(population[p]);
population.RemoveAt(p);
}
log.WriteLine("Sampling " + samples.Count + " out of " + ranges.Count + " shards (" + (100.0 * samples.Count / ranges.Count).ToString("N1") + "%) ...");
}
log.WriteLine();
const string FORMAT_STRING = "{0,9} ║{1,10}{6,6} {2,-29} ║{3,10}{7,7} {4,-37} ║{5,10}";
const string SCALE_KEY = "....--------========########M";
const string SCALE_VAL = "....--------========########@@@@@@@@M";
log.WriteLine(FORMAT_STRING, "Count", "Keys", SCALE_KEY, "Values", SCALE_VAL, "Total", "med.", "med.");
var rangeOptions = new FdbRangeOptions { Mode = FdbStreamingMode.WantAll };
samples = samples.OrderBy(x => x.Begin).ToList();
long globalSize = 0;
long globalCount = 0;
int workers = 8; // Math.Max(4, Environment.ProcessorCount);
var sw = Stopwatch.StartNew();
var tasks = new List<Task>();
int n = samples.Count;
while (samples.Count > 0)
{
while (tasks.Count < workers && samples.Count > 0)
{
var range = samples[0];
samples.RemoveAt(0);
tasks.Add(Task.Run(async () =>
{
var kk = new RobustHistogram(RobustHistogram.TimeScale.Ticks);
var vv = new RobustHistogram(RobustHistogram.TimeScale.Ticks);
#region Method 1: get_range everything...
using (var tr = db.BeginTransaction(ct))
{
long keySize = 0;
long valueSize = 0;
long count = 0;
int iter = 0;
var beginSelector = FdbKeySelector.FirstGreaterOrEqual(range.Begin);
var endSelector = FdbKeySelector.FirstGreaterOrEqual(range.End);
while (true)
{
FdbRangeChunk data = default(FdbRangeChunk);
FdbException error = null;
try
{
data = await tr.Snapshot.GetRangeAsync(
beginSelector,
endSelector,
rangeOptions,
iter
).ConfigureAwait(false);
}
catch (FdbException e)
{
error = e;
}
if (error != null)
{
await tr.OnErrorAsync(error.Code).ConfigureAwait(false);
continue;
}
if (data.Count == 0) break;
count += data.Count;
foreach (var kvp in data.Chunk)
{
keySize += kvp.Key.Count;
valueSize += kvp.Value.Count;
kk.Add(TimeSpan.FromTicks(kvp.Key.Count));
vv.Add(TimeSpan.FromTicks(kvp.Value.Count));
}
if (!data.HasMore) break;
beginSelector = FdbKeySelector.FirstGreaterThan(data.Last.Key);
++iter;
}
long totalSize = keySize + valueSize;
Interlocked.Add(ref globalSize, totalSize);
Interlocked.Add(ref globalCount, count);
lock (log)
{
log.WriteLine(FORMAT_STRING, count.ToString("N0"), FormatSize(keySize), kk.GetDistribution(begin: 1, end: 12000, fold: 2), FormatSize(valueSize), vv.GetDistribution(begin: 1, end: 120000, fold: 2), FormatSize(totalSize), FormatSize((int)Math.Ceiling(kk.Median)), FormatSize((int)Math.Ceiling(vv.Median)));
}
}
#endregion
#region Method 2: estimate the count using key selectors...
//long counter = await Fdb.System.EstimateCountAsync(db, range, ct);
//Console.WriteLine("COUNT = " + counter.ToString("N0"));
#endregion
}, ct));
}
var done = await Task.WhenAny(tasks);
tasks.Remove(done);
}
await Task.WhenAll(tasks);
sw.Stop();
log.WriteLine();
if (n != ranges.Count)
{
log.WriteLine("Sampled " + FormatSize(globalSize) + " (" + globalSize.ToString("N0") + " bytes) and " + globalCount.ToString("N0") + " keys in " + sw.Elapsed.TotalSeconds.ToString("N1") + " sec");
log.WriteLine("> Estimated total size is " + FormatSize(globalSize * ranges.Count / n));
}
else
{
log.WriteLine("Found " + FormatSize(globalSize) + " (" + globalSize.ToString("N0") + " bytes) and " + globalCount.ToString("N0") + " keys in " + sw.Elapsed.TotalSeconds.ToString("N1") + " sec");
// compare to the whole cluster
ranges = await Fdb.System.GetChunksAsync(db, FdbKey.MinValue, FdbKey.MaxValue, ct);
log.WriteLine("> This directory contains ~{0:N2}% of all data", (100.0 * n / ranges.Count));
}
log.WriteLine();
}
