// Copyright (C) 2025, The Duplicati Team // https://duplicati.com, hello@duplicati.com // // Permission is hereby granted, free of charge, to any person obtaining a // copy of this software and associated documentation files (the "Software"), // to deal in the Software without restriction, including without limitation // the rights to use, copy, modify, merge, publish, distribute, sublicense, // and/or sell copies of the Software, and to permit persons to whom the // Software is furnished to do so, subject to the following conditions: // // The above copyright notice and this permission notice shall be included in // all copies or substantial portions of the Software. // // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS // OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING // FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER // DEALINGS IN THE SOFTWARE. using System; using System.Buffers; using System.Collections.Concurrent; using System.Collections.Generic; using System.Data; using System.Diagnostics; using System.Linq; using System.Threading; using System.Threading.Tasks; using CoCoL; using Duplicati.Library.Main.Database; using Microsoft.Extensions.Caching.Memory; namespace Duplicati.Library.Main.Operation.Restore { /// /// Process that manages the block requests and responses to/from the /// `FileProcessor` process by caching the blocks. /// internal class BlockManager { /// /// The log tag for this class. /// private static readonly string LOGTAG = Logging.Log.LogTagFromType(); /// /// Dictionary for data blocks that are being cached. Whenever a block /// is requested, it checks if it is in the cache. If it is not, it /// will request the block from the corresponding volume. The requester /// will be given a `Task` that will be completed when the block is /// available. The cache will also keep track of how many times a block /// is requested, and only remove it from the cache when all requests /// have been fulfilled. This is to ensure that the cache is not /// prematurely evicted, while keeping the memory usage low. The /// dictionary also keeps track of how many readers are accessing it, /// and will retire the volume request channel when the last reader is /// done. Once disposed, the dictionary will clean up the database /// table that was used to keep track of the block counts. /// internal class SleepableDictionary : IDisposable { /// /// Channel for submitting block requests from a volume. /// private readonly IWriteChannel m_volume_request; /// /// The dictionary holding the cached blocks. /// private readonly MemoryCache m_block_cache; /// /// The dictionary holding the `Task` for each block request in flight. /// private readonly ConcurrentDictionary> m_waiters = new(); /// /// The number of readers accessing this dictionary. Used during shutdown / cleanup. /// private int readers = 0; /// /// Internal stopwatch for profiling the cache eviction. /// private readonly Stopwatch sw_cacheevict; /// /// Internal stopwatch for profiling the `CheckCounts` method. /// private readonly Stopwatch sw_checkcounts; /// /// Internal stopwatch for profiling setting up the waiters. /// private readonly Stopwatch sw_get_wait; /// /// Dictionary for keeping track of how many times each block is requested. Used to determine when a block is no longer needed. /// private readonly Dictionary m_blockcount = new(); /// /// Lock for the block count dictionary. /// private readonly object m_blockcount_lock = new(); /// /// Dictionary for keeping track of how many times each volume is requested. Used to determine when a volume is no longer needed. /// private readonly Dictionary m_volumecount = new(); /// /// The options for the restore. /// private readonly Options m_options; /// /// The cache eviction options. Used for registering a callback when a block is evicted from the cache. /// private readonly MemoryCacheEntryOptions m_entry_options = new(); /// /// Dictionary to keep track of how many active readers are accessing each block. On eviction, the byte[] buffer can only be returned to the ArrayPool if there are no active readers. /// private readonly ConcurrentDictionary m_active_readers = []; /// /// Initializes a new instance of the class. /// /// The database holding information about how many of each block this restore requires. /// Channel for submitting block requests from a volume. /// Number of readers accessing this dictionary. Used during shutdown / cleanup. public SleepableDictionary(LocalRestoreDatabase db, IWriteChannel volume_request, Options options, int readers) { m_options = options; m_volume_request = volume_request; var cache_options = new MemoryCacheOptions(); m_block_cache = new MemoryCache(cache_options); m_entry_options.RegisterPostEvictionCallback(async (key, value, reason, state) => { bool was_present = false; while (true) { lock (m_blockcount_lock) { if (m_active_readers.TryGetValue((long)key, out var ac)) { if (ac == 0) { m_block_cache.Remove(key); was_present = true; break; } } else break; } await Task.Delay(10).ConfigureAwait(false); } if (was_present) ArrayPool.Shared.Return((byte[])value); }); this.readers = readers; sw_cacheevict = options.InternalProfiling ? new() : null; sw_checkcounts = options.InternalProfiling ? new() : null; sw_get_wait = options.InternalProfiling ? new() : null; foreach (var (block_id, volume_id) in db.GetBlocksAndVolumeIDs(options.SkipMetadata)) { var bc = m_blockcount.TryGetValue(block_id, out var c); m_blockcount[block_id] = bc ? c + 1 : 1; var vc = m_volumecount.TryGetValue(volume_id, out var v); m_volumecount[volume_id] = vc ? v + 1 : 1; } } /// /// Decrements the block counters for the block and volume, and checks /// if the block is still needed. If the block is no longer needed, it /// will be removed from the cache. If the volume is no longer needed, /// the VolumeDownloader will be notified. /// /// The block request to check. public async Task CheckCounts(BlockRequest blockRequest) { long error_block_id = -1; long error_volume_id = -1; var emit_evict = false; byte[] data = null; lock (m_blockcount_lock) { sw_checkcounts?.Start(); var active = m_active_readers.TryGetValue(blockRequest.BlockID, out var ac) ? ac - 1 : 0; if (active == 0) m_active_readers.Remove(blockRequest.BlockID, out var _); else m_active_readers[blockRequest.BlockID] = active; var block_count = m_blockcount.TryGetValue(blockRequest.BlockID, out var c) ? c - 1 : 0; if (block_count > 0) { m_blockcount[blockRequest.BlockID] = block_count; } else if (block_count == 0) { // Evict the block from the cache and check if the volume is no longer needed. m_blockcount.Remove(blockRequest.BlockID); data = m_block_cache.Get(blockRequest.BlockID); m_block_cache.Remove(blockRequest.BlockID); } else // block_count < 0 { error_block_id = blockRequest.BlockID; } var vol_count = m_volumecount.TryGetValue(blockRequest.VolumeID, out var vc) ? vc - 1 : 0; if (vol_count > 0) { m_volumecount[blockRequest.VolumeID] = vol_count; } else if (vol_count == 0) { m_volumecount.Remove(blockRequest.VolumeID); blockRequest.CacheDecrEvict = true; emit_evict = true; } else // vol_count < 0 { error_volume_id = blockRequest.VolumeID; } sw_checkcounts?.Stop(); } if (data != null) ArrayPool.Shared.Return(data); // Notify the `VolumeManager` that it should evict the volume. if (emit_evict) await m_volume_request.WriteAsync(blockRequest).ConfigureAwait(false); if (error_block_id != -1) { Logging.Log.WriteWarningMessage(LOGTAG, "BlockCountError", null, $"Block {blockRequest.BlockID} has a count below 0"); } if (error_volume_id != -1) { Logging.Log.WriteWarningMessage(LOGTAG, "VolumeCountError", null, $"Volume {blockRequest.VolumeID} has a count below 0"); } } /// /// Get a block from the cache. If the block is not in the cache, /// it will request the block from the volume and return a `Task` /// that will be completed when the block is available. /// /// The requested block. /// A `Task` holding the data block. public Task Get(BlockRequest block_request) { lock (m_blockcount_lock) { m_active_readers[block_request.BlockID] = m_active_readers.TryGetValue(block_request.BlockID, out var c) ? c + 1 : 1; } // Check if the block is already in the cache, and return it if it is. if (m_block_cache != null && m_block_cache.TryGetValue(block_request.BlockID, out byte[] value)) { return Task.FromResult(value); } // If the block is not in the cache, request it from the volume. sw_get_wait?.Start(); var tcs = new TaskCompletionSource(); var new_tcs = m_waiters.GetOrAdd(block_request.BlockID, tcs); if (tcs == new_tcs) { // We are the first to request this block m_volume_request.Write(block_request); } sw_get_wait?.Stop(); return new_tcs.Task; } /// /// Set a block in the cache. If the block is already in the cache, /// it will be replaced. If the block is not in the cache, it will /// be added. If the block is no longer needed, it will be removed /// from the cache. If the block is requested while it is being /// removed, the requester will be given a `Task` that will be /// completed when the block is available. /// /// The block request related to the value. /// The byte[] buffer holding the block data. public void Set(BlockRequest blockRequest, byte[] value) { m_block_cache.Set(blockRequest.BlockID, value); // Notify any waiters that the block is available. if (m_waiters.TryRemove(blockRequest.BlockID, out var tcs)) { tcs.SetResult(value); } sw_cacheevict?.Start(); if (m_block_cache.Count > m_options.RestoreCacheMax) { m_block_cache.Compact(m_options.RestoreCacheMax == 0 ? 1.0 : m_options.RestoreCacheEvict); } sw_cacheevict?.Stop(); } /// /// Retire the dictionary. This will decrement the number of readers /// accessing the dictionary, and if there are no more readers, it /// will retire the volume request channel, effectively shutting /// down the restore process network. /// public void Retire() { if (Interlocked.Decrement(ref readers) <= 0) { m_volume_request.Retire(); } } /// /// Clean up the dictionary. This will remove the database table /// that was used to keep track of the block counts. /// public void Dispose() { // Verify that the tables are empty var blockcount = m_blockcount.Sum(x => x.Value); var volumecount = m_volumecount.Sum(x => x.Value); if (blockcount != 0) { var blocks = m_blockcount.Where(x => x.Value != 0).Select(x => x.Key).ToArray(); var blockids = string.Join(", ", blocks); Logging.Log.WriteWarningMessage(LOGTAG, "BlockCountError", null, $"Block count in SleepableDictionarys block table is not zero: {blockcount}{Environment.NewLine}"); } if (m_active_readers.Count > 0) { Logging.Log.WriteWarningMessage(LOGTAG, "BlockCountError", null, $"There are still {m_active_readers.Count} files being read by {m_active_readers.Sum(x => x.Value)} readers"); } if (volumecount != 0) { var vols = m_volumecount.Where(x => x.Value != 0).Select(x => x.Key).ToArray(); var volids = string.Join(", ", vols); Logging.Log.WriteWarningMessage(LOGTAG, "VolumeCountError", null, $"Volume count in SleepableDictionarys volume table is not zero: {volumecount}{Environment.NewLine}Volumes: {volids}"); } if (m_options.InternalProfiling) { Logging.Log.WriteProfilingMessage(LOGTAG, "InternalTimings", $"Sleepable dictionary - CheckCounts: {sw_checkcounts.ElapsedMilliseconds}ms, Get wait: {sw_get_wait.ElapsedMilliseconds}ms, Cache evict: {sw_cacheevict.ElapsedMilliseconds}ms"); } if (m_block_cache.Count > 0) { Logging.Log.WriteWarningMessage(LOGTAG, "BlockCacheMismatch", null, $"Internal Block cache is not empty: {m_block_cache.Count}"); Logging.Log.WriteWarningMessage(LOGTAG, "BlockCacheMismatch", null, $"Block counts in cache ({m_blockcount.Count}): {string.Join(", ", m_blockcount.Select(x => x.Value))}"); } } /// /// Cancel all pending requests. This will set an exception on all /// pending requests, effectively cancelling them. /// public void CancelAll() { foreach (var tcs in m_waiters.Values) { tcs.SetException(new RetiredException("Request waiter")); } } } /// /// Run the block manager process. This will create a cache for the /// blocks, and start two tasks: one for reading blocks from the input /// channel (data blocks from the volumes) and storing them in the /// cache, and one for reading block requests from the `FileProcessor`, /// accessing the cache for the blocks, and writing the resulting /// blocks back to the `FileProcessor`. /// /// The named channels for the restore operation. /// The database holding information about how many of each block this restore requires. /// The restore options. /// The channels for reading block requests from the `FileProcessor`. /// The channels for writing block responses back to the `FileProcessor`. public static Task Run(Channels channels, LocalRestoreDatabase db, Options options, IChannel[] fp_requests, IChannel[] fp_responses) { return AutomationExtensions.RunTask( new { Input = channels.DecompressedBlock.AsRead(), Output = channels.VolumeRequestResponse.AsWrite() }, async self => { // Create a cache for the blocks, using SleepableDictionary cache = new(db, self.Output, options, fp_requests.Length); // The volume consumer will read blocks from the input channel (data blocks from the volumes) and store them in the cache. var volume_consumer = Task.Run(async () => { Stopwatch sw_read = options.InternalProfiling ? new() : null; Stopwatch sw_set = options.InternalProfiling ? new() : null; try { while (true) { sw_read?.Start(); var (block_request, data) = await self.Input.ReadAsync().ConfigureAwait(false); sw_read?.Stop(); sw_set?.Start(); cache.Set(block_request, data); sw_set?.Stop(); } } catch (RetiredException) { Logging.Log.WriteVerboseMessage(LOGTAG, "RetiredProcess", null, "BlockManager Volume consumer retired"); if (options.InternalProfiling) { Logging.Log.WriteProfilingMessage(LOGTAG, "InternalTimings", $"Volume consumer - Read: {sw_read.ElapsedMilliseconds}ms, Set: {sw_set.ElapsedMilliseconds}ms"); } // Cancel any remaining readers - although there shouldn't be any. cache.CancelAll(); } catch (Exception ex) { Logging.Log.WriteWarningMessage(LOGTAG, "VolumeConsumerError", ex, "Error in volume consumer"); // Cancel any remaining readers - although there shouldn't be any. cache.CancelAll(); } }); // The block handlers will read block requests from the `FileProcessor`, access the cache for the blocks, and write the resulting blocks to the `FileProcessor`. var block_handlers = fp_requests.Zip(fp_responses, (req, res) => Task.Run(async () => { Stopwatch sw_req = options.InternalProfiling ? new() : null; Stopwatch sw_resp = options.InternalProfiling ? new() : null; Stopwatch sw_cache = options.InternalProfiling ? new() : null; Stopwatch sw_get = options.InternalProfiling ? new() : null; try { while (true) { sw_req?.Start(); var block_request = await req.ReadAsync().ConfigureAwait(false); sw_req?.Stop(); if (block_request.CacheDecrEvict) { sw_cache?.Start(); // Target file already had the block. await cache.CheckCounts(block_request).ConfigureAwait(false); sw_cache?.Stop(); } else { sw_get?.Start(); var data = await cache.Get(block_request).ConfigureAwait(false); sw_get?.Stop(); sw_resp?.Start(); await res.WriteAsync(data).ConfigureAwait(false); sw_resp?.Stop(); } } } catch (RetiredException) { Logging.Log.WriteVerboseMessage(LOGTAG, "RetiredProcess", null, "BlockManager Block handler retired"); if (options.InternalProfiling) { Logging.Log.WriteProfilingMessage(LOGTAG, "InternalTimings", $"Block handler - Req: {sw_req.ElapsedMilliseconds}ms, Resp: {sw_resp.ElapsedMilliseconds}ms, Cache: {sw_cache.ElapsedMilliseconds}ms, Get: {sw_get.ElapsedMilliseconds}ms"); } cache.Retire(); } })).ToArray(); await Task.WhenAll([volume_consumer, .. block_handlers]).ConfigureAwait(false); }); } } }