private SelfCheckState GetSelfCheckState(Context context)
{
if (!_fileSystem.FileExists(_selfCheckFilePath))
{
return(SelfCheckState.OutOfDate);
}
var content = _fileSystem.ReadAllText(_selfCheckFilePath);
_tracer.Debug(context, $"SelfCheck: got self check status from disk '{content}'.");
return(SelfCheckState.TryParse(content) ?? SelfCheckState.OutOfDate);
}
private void UpdateSelfCheckStateOnDisk(Context context, SelfCheckState selfCheckState)
{
try
{
_tracer.Debug(context, $"SelfCheck: Updating self check status on disk with new state: {selfCheckState.ToParseableString()}");
_fileSystem.WriteAllText(_selfCheckFilePath, selfCheckState.ToParseableString());
}
catch (IOException e)
{
_tracer.Warning(context, $"SelfCheck: Failed updating self check status on disk: {e}");
}
}
public void SelfCheckStateTests()
{
// Reparsing date time to loose precision in order for equality to work.
var now = DateTimeUtilities.FromReadableTimestamp(DateTime.UtcNow.ToReadableString()).Value;
var state1 = new SelfCheckState("Epoch1", now, ContentHash.Random());
var reparsedState1 = SelfCheckState.TryParse(state1.ToParseableString());
reparsedState1.Should().NotBeNull();
reparsedState1.Value.Should().Be(state1);
var state2 = new SelfCheckState("Epoch1", now);
var reparsedState2 = SelfCheckState.TryParse(state2.ToParseableString());
reparsedState2.Should().NotBeNull();
reparsedState2.Value.Should().Be(state2);
}
private async Task <Result <SelfCheckResult> > SelfCheckContentDirectoryCoreAsync(
OperationContext context,
SelfCheckState selfCheckState,
SelfCheckStatus status)
{
_tracer.Always(context, "Starting self check.");
// Self checking procedure validates that in-memory content directory
// is valid in respect to the state on disk.
// Namely, it checks that the hashes for all the files and their size are correct.
var stopwatch = Stopwatch.StartNew();
// Enumerating files from disk instead of looking them up from content directory.
// This is done due to simplicity (we don't have to worry about replicas) and because an additional IO cost is negligible compared to the cost of rehashing.
var contentHashes = _contentStoreInternal.ReadSnapshotFromDisk(context).ListOrderedByHash();
_tracer.Debug(context, $"SelfCheck: Enumerated {contentHashes.Count} entries from disk by {stopwatch.ElapsedMilliseconds}ms.");
stopwatch.Restart();
// Trying to restore the index of a hash that we processed before.
int index = 0;
if (status == SelfCheckStatus.InProgress && selfCheckState.LastPosition != null)
{
index = findNextIndexToProcess(selfCheckState.LastPosition.Value);
_tracer.Debug(context, $"SelfCheck: skipping {index} elements based on previous state '{selfCheckState.ToParseableString()}'.");
}
else
{
string statusAsString = status == SelfCheckStatus.Force ? "the epoch has changed" : "a self check is out of date";
_tracer.Debug(
context,
$"SelfCheck: starting self check for the entire content directory because {statusAsString}. Previous state '{selfCheckState.ToParseableString()}'.");
}
// Time span for tracking progress.
TimeSpan progressTracker = TimeSpan.FromSeconds(0);
long processedBytes = 0;
int invalidEntries = 0;
int processedEntries = 0;
for (; index < contentHashes.Count; index++)
{
if (context.Token.IsCancellationRequested)
{
_tracer.Debug(context, "SelfCheck: Exiting self check because cancellation was requested.");
break;
}
var hashInfo = contentHashes[index];
processedEntries++;
var(isValid, error) = await ValidateFileAsync(context, hashInfo.Hash, hashInfo.Payload);
if (!isValid)
{
_tracer.Warning(context, $"SelfCheck: Found invalid entry in cache. Hash={hashInfo.Hash.ToShortString()}. {error}. Evicting the file...");
await _contentStoreInternal.RemoveInvalidContentAsync(context, hashInfo.Hash);
invalidEntries++;
}
// Tracking the progress if needed.
traceProgressIfNeeded(hashInfo.Hash, index);
// If the current entry is not the last one, and we reached the number of invalid files,
// then exiting the loop.
if (invalidEntries == _settings.InvalidFilesLimit && index != contentHashes.Count - 1)
{
_tracer.Debug(context, $"SelfCheck: Exiting self check because invalid file limit of {_settings.InvalidFilesLimit} is reached.");
break;
}
if (_settings.HashAnalysisDelay != null)
{
await Task.Delay(_settings.HashAnalysisDelay.Value, context.Token);
}
}
if (index == contentHashes.Count)
{
// All the items are processed. Saving new stable checkpoint
UpdateSelfCheckStateOnDisk(context, SelfCheckState.SelfCheckComplete(_settings.Epoch, _clock.UtcNow));
}
else
{
// The loop was interrupted. Saving an incremental state.
var newStatus = selfCheckState.WithEpochAndPosition(_settings.Epoch, contentHashes[index].Hash);
UpdateSelfCheckStateOnDisk(context, newStatus);
}
return(Result.Success(new SelfCheckResult(invalidHashes: invalidEntries, totalProcessedFiles: processedEntries)));
int findNextIndexToProcess(ContentHash lastProcessedHash)
{
var binarySearchResult = contentHashes.BinarySearch(new PayloadFromDisk <FileInfo>(lastProcessedHash, default), new ByHashPayloadFromDiskComparer <FileInfo>());
int targetIndex = 0;
if (binarySearchResult >= 0 && binarySearchResult < contentHashes.Count - 1)
{
targetIndex = binarySearchResult + 1;
}
else
{
// The exact match is not found (which is fine, because the state on disk may changed between app invocations).
// BinarySearch returns a negative value that bitwise complement of the closest element in the sorted array.
binarySearchResult = ~binarySearchResult;
if (binarySearchResult < contentHashes.Count)
{
targetIndex = binarySearchResult;
}
}
return(targetIndex);
}
void traceProgressIfNeeded(ContentHash currentHash, int currentHashIndex)
{
processedBytes += contentHashes[currentHashIndex].Payload.Length;
var swTime = stopwatch.Elapsed;
if (swTime - progressTracker > _settings.ProgressReportingInterval)
{
// It is possible to have multiple replicas with the same hash.
// We need to save the state only when *all* the replicas are processed.
// So we check if the next item has a different hash.
// No check is performed on the last element because the state will be saved immediately after
if (currentHashIndex + 1 < contentHashes.Count && currentHash != contentHashes[currentHashIndex + 1].Hash)
{
var speed = ((double)processedBytes / (1024 * 1024)) / _settings.ProgressReportingInterval.TotalSeconds;
_tracer.Always(context, $"SelfCheck: processed {index}/{contentHashes.Count}: {new SelfCheckResult(invalidEntries, processedEntries)}. Hashing speed {speed:0.##}Mb/s.");
processedBytes = 0;
// Saving incremental state
var newStatus = selfCheckState.WithEpochAndPosition(_settings.Epoch, currentHash);
UpdateSelfCheckStateOnDisk(context, newStatus);
progressTracker = swTime;
}
}
}
}
