feat: add SequenceSet for sparse deletion tracking with secure erase (Gap 1.7)
Replace HashSet<ulong> _deleted in MsgBlock with SequenceSet — a sorted-range
list that compresses contiguous deletions into (Start, End) intervals. Adds
O(log n) Contains/Add via binary search on range count, matching Go's avl.SequenceSet
semantics with a simpler implementation.
- Add SequenceSet.cs: sorted-range compressed set with Add/Remove/Contains/Count/Clear
and IEnumerable<ulong> in ascending order. Binary search for all O(log n) ops.
- Replace HashSet<ulong> _deleted and _skipSequences in MsgBlock with SequenceSet.
- Add secureErase parameter (default false) to MsgBlock.Delete(): when true, payload
bytes are overwritten with RandomNumberGenerator.Fill() before the delete record is
written, making original content unrecoverable on disk.
- Update FileStore.DeleteInBlock() to propagate secureErase flag.
- Update FileStore.EraseMsg() to use secureErase: true via block layer instead of
delegating to RemoveMsg().
- Add SequenceSetTests.cs: 25 tests covering Add, Remove, Contains, Count, range
compression, gap filling, bridge merges, enumeration, boundary values, round-trip.
- Add FileStoreTombstoneTrackingTests.cs: 12 tests covering SequenceSet tracking in
MsgBlock, tombstone persistence through RebuildIndex recovery, secure erase
payload overwrite verification, and FileStore.EraseMsg integration.
Go reference: filestore.go:5267 (removeMsg), filestore.go:5890 (eraseMsg),
avl/seqset.go (SequenceSet).
This commit is contained in:
@@ -732,14 +732,17 @@ public sealed class FileStore : IStreamStore, IAsyncDisposable, IDisposable
|
||||
|
||||
/// <summary>
|
||||
/// Soft-deletes a message in the block that contains it.
|
||||
/// When <paramref name="secureErase"/> is <c>true</c>, payload bytes are
|
||||
/// overwritten with random data before the delete record is written.
|
||||
/// Reference: golang/nats-server/server/filestore.go:5890 (eraseMsg).
|
||||
/// </summary>
|
||||
private void DeleteInBlock(ulong sequence)
|
||||
private void DeleteInBlock(ulong sequence, bool secureErase = false)
|
||||
{
|
||||
foreach (var block in _blocks)
|
||||
{
|
||||
if (sequence >= block.FirstSequence && sequence <= block.LastSequence)
|
||||
{
|
||||
block.Delete(sequence);
|
||||
block.Delete(sequence, secureErase);
|
||||
return;
|
||||
}
|
||||
}
|
||||
@@ -1386,15 +1389,25 @@ public sealed class FileStore : IStreamStore, IAsyncDisposable, IDisposable
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Overwrites a message with zeros and then soft-deletes it.
|
||||
/// Secure-erases a message: overwrites its payload bytes with random data on disk,
|
||||
/// then soft-deletes it (same in-memory semantics as <see cref="RemoveMsg"/>).
|
||||
/// Returns <c>true</c> if the sequence existed and was erased.
|
||||
/// Reference: golang/nats-server/server/filestore.go — EraseMsg.
|
||||
/// Reference: golang/nats-server/server/filestore.go:5890 (eraseMsg).
|
||||
/// </summary>
|
||||
public bool EraseMsg(ulong seq)
|
||||
{
|
||||
// In .NET we don't do physical overwrite — just remove from the in-memory
|
||||
// cache and soft-delete in the block layer (same semantics as RemoveMsg).
|
||||
return RemoveMsg(seq);
|
||||
if (!_messages.Remove(seq, out _))
|
||||
return false;
|
||||
|
||||
if (_messages.Count == 0)
|
||||
_first = _last + 1;
|
||||
else
|
||||
_first = _messages.Keys.Min();
|
||||
|
||||
// Secure erase: overwrite payload bytes with random data before marking deleted.
|
||||
// Reference: golang/nats-server/server/filestore.go:5890 (eraseMsg).
|
||||
DeleteInBlock(seq, secureErase: true);
|
||||
return true;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
|
||||
@@ -10,6 +10,7 @@
|
||||
// sequentially as binary records (using MessageRecord). Blocks are sealed
|
||||
// (read-only) when they reach a configurable size limit.
|
||||
|
||||
using System.Security.Cryptography;
|
||||
using Microsoft.Win32.SafeHandles;
|
||||
|
||||
namespace NATS.Server.JetStream.Storage;
|
||||
@@ -25,10 +26,14 @@ public sealed class MsgBlock : IDisposable
|
||||
private readonly FileStream _file;
|
||||
private readonly SafeFileHandle _handle;
|
||||
private readonly Dictionary<ulong, (long Offset, int Length)> _index = new();
|
||||
private readonly HashSet<ulong> _deleted = new();
|
||||
// Go: msgBlock.dmap — avl.SequenceSet for sparse deletion tracking.
|
||||
// Reference: golang/nats-server/server/avl/seqset.go (SequenceSet).
|
||||
// .NET uses a sorted-range list (see SequenceSet.cs) for O(log n) ops with
|
||||
// range compression for contiguous deletion runs (TTL, bulk remove).
|
||||
private readonly SequenceSet _deleted = new();
|
||||
// Go: SkipMsg writes tombstone records with empty subject — tracked separately so
|
||||
// recovery can distinguish intentional sequence gaps from soft-deleted messages.
|
||||
private readonly HashSet<ulong> _skipSequences = new();
|
||||
private readonly SequenceSet _skipSequences = new();
|
||||
private readonly long _maxBytes;
|
||||
private readonly ReaderWriterLockSlim _lock = new();
|
||||
private long _writeOffset; // Tracks the append position independently of FileStream.Position
|
||||
@@ -95,7 +100,7 @@ public sealed class MsgBlock : IDisposable
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>Count of soft-deleted messages.</summary>
|
||||
/// <summary>Count of soft-deleted messages. Mirrors Go's msgBlock.dmap.Size().</summary>
|
||||
public ulong DeletedCount
|
||||
{
|
||||
get
|
||||
@@ -357,10 +362,18 @@ public sealed class MsgBlock : IDisposable
|
||||
/// Soft-deletes a message by sequence number. Re-encodes the record on disk
|
||||
/// with the deleted flag set (and updated checksum) so the deletion survives recovery.
|
||||
/// Also evicts the sequence from the write cache.
|
||||
/// When <paramref name="secureErase"/> is <c>true</c>, the payload bytes inside
|
||||
/// the encoded record are overwritten with cryptographically random data before
|
||||
/// the record is re-written — ensuring the original payload is unrecoverable.
|
||||
/// Reference: golang/nats-server/server/filestore.go:5890 (eraseMsg).
|
||||
/// </summary>
|
||||
/// <param name="sequence">The sequence number to delete.</param>
|
||||
/// <param name="secureErase">
|
||||
/// When <c>true</c>, payload bytes are filled with random data before the
|
||||
/// record is written back. Defaults to <c>false</c>.
|
||||
/// </param>
|
||||
/// <returns>True if the message was deleted; false if already deleted or not found.</returns>
|
||||
public bool Delete(ulong sequence)
|
||||
public bool Delete(ulong sequence, bool secureErase = false)
|
||||
{
|
||||
_lock.EnterWriteLock();
|
||||
try
|
||||
@@ -377,12 +390,23 @@ public sealed class MsgBlock : IDisposable
|
||||
RandomAccess.Read(_handle, buffer, entry.Offset);
|
||||
var record = MessageRecord.Decode(buffer);
|
||||
|
||||
ReadOnlyMemory<byte> payload = record.Payload;
|
||||
if (secureErase && payload.Length > 0)
|
||||
{
|
||||
// Go: eraseMsg — overwrite payload region with random bytes so the
|
||||
// original content is unrecoverable from disk.
|
||||
// Reference: golang/nats-server/server/filestore.go:5890 (eraseMsg).
|
||||
var randomPayload = new byte[payload.Length];
|
||||
RandomNumberGenerator.Fill(randomPayload);
|
||||
payload = randomPayload;
|
||||
}
|
||||
|
||||
var deletedRecord = new MessageRecord
|
||||
{
|
||||
Sequence = record.Sequence,
|
||||
Subject = record.Subject,
|
||||
Headers = record.Headers,
|
||||
Payload = record.Payload,
|
||||
Payload = payload,
|
||||
Timestamp = record.Timestamp,
|
||||
Deleted = true,
|
||||
};
|
||||
@@ -498,13 +522,24 @@ public sealed class MsgBlock : IDisposable
|
||||
get
|
||||
{
|
||||
_lock.EnterReadLock();
|
||||
try { return _skipSequences.Count > 0 ? _skipSequences.Max() : 0UL; }
|
||||
try
|
||||
{
|
||||
if (_skipSequences.IsEmpty)
|
||||
return 0UL;
|
||||
// SequenceSet enumerates in ascending order; last element is max.
|
||||
ulong max = 0;
|
||||
foreach (var seq in _skipSequences)
|
||||
max = seq; // last wins since iteration is ascending
|
||||
return max;
|
||||
}
|
||||
finally { _lock.ExitReadLock(); }
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Exposes the set of soft-deleted sequence numbers for read-only inspection.
|
||||
/// Returns a snapshot as a <see cref="HashSet{T}"/> so callers can use
|
||||
/// standard <see cref="IReadOnlySet{T}"/> operations.
|
||||
/// Reference: golang/nats-server/server/filestore.go — dmap access for state queries.
|
||||
/// </summary>
|
||||
public IReadOnlySet<ulong> DeletedSequences
|
||||
@@ -512,7 +547,7 @@ public sealed class MsgBlock : IDisposable
|
||||
get
|
||||
{
|
||||
_lock.EnterReadLock();
|
||||
try { return new HashSet<ulong>(_deleted); }
|
||||
try { return _deleted.ToHashSet(); }
|
||||
finally { _lock.ExitReadLock(); }
|
||||
}
|
||||
}
|
||||
|
||||
230
src/NATS.Server/JetStream/Storage/SequenceSet.cs
Normal file
230
src/NATS.Server/JetStream/Storage/SequenceSet.cs
Normal file
@@ -0,0 +1,230 @@
|
||||
// Reference: golang/nats-server/server/avl/seqset.go
|
||||
// Go uses an AVL tree with bitmask nodes (2048 sequences per node).
|
||||
// .NET port uses a sorted list of (Start, End) ranges — simpler, still O(log n)
|
||||
// via binary search for the common case of mostly-sequential deletions in JetStream.
|
||||
//
|
||||
// Range compression: adding sequences 1, 2, 3 stores as a single range [1, 3].
|
||||
// This is significantly more memory efficient than HashSet<ulong> for contiguous
|
||||
// deletion runs (e.g. TTL expiry, bulk removes), which are the dominant pattern.
|
||||
//
|
||||
// Not thread-safe — callers must hold their own lock (MsgBlock uses ReaderWriterLockSlim).
|
||||
|
||||
namespace NATS.Server.JetStream.Storage;
|
||||
|
||||
/// <summary>
|
||||
/// A memory-efficient sparse set for storing unsigned sequence numbers, using
|
||||
/// range compression to merge contiguous sequences into (Start, End) intervals.
|
||||
///
|
||||
/// Analogous to Go's <c>avl.SequenceSet</c> but implemented with a sorted list
|
||||
/// of ranges for simplicity. Binary search gives O(log n) Contains/Add on the
|
||||
/// number of distinct ranges (not the total count of sequences).
|
||||
///
|
||||
/// Reference: golang/nats-server/server/avl/seqset.go (SequenceSet struct).
|
||||
/// </summary>
|
||||
internal sealed class SequenceSet : IEnumerable<ulong>
|
||||
{
|
||||
// Sorted list of non-overlapping, non-adjacent ranges in ascending order.
|
||||
// Invariant: for all i, ranges[i].End + 1 < ranges[i+1].Start (strict gap between consecutive ranges).
|
||||
private readonly List<(ulong Start, ulong End)> _ranges = [];
|
||||
|
||||
/// <summary>Total number of sequences across all ranges.</summary>
|
||||
public int Count
|
||||
{
|
||||
get
|
||||
{
|
||||
var total = 0;
|
||||
foreach (var (start, end) in _ranges)
|
||||
total += (int)(end - start + 1);
|
||||
return total;
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>True when the set contains no sequences.</summary>
|
||||
public bool IsEmpty => _ranges.Count == 0;
|
||||
|
||||
/// <summary>
|
||||
/// Adds <paramref name="seq"/> to the set.
|
||||
/// Merges adjacent or overlapping ranges automatically.
|
||||
/// Returns <c>true</c> if the sequence was not already present.
|
||||
/// Reference: golang/nats-server/server/avl/seqset.go:44 (Insert).
|
||||
/// </summary>
|
||||
public bool Add(ulong seq)
|
||||
{
|
||||
// Strategy: find the position where seq belongs (binary search by Start),
|
||||
// then check if seq is already in the previous range or the range at that position,
|
||||
// and determine left/right adjacency for merging.
|
||||
|
||||
// Binary search: find the first index i where _ranges[i].Start > seq.
|
||||
// Everything before i has Start <= seq. The range that might contain seq
|
||||
// is at i-1 (if it exists and its End >= seq).
|
||||
var lo = 0;
|
||||
var hi = _ranges.Count; // exclusive upper bound
|
||||
while (lo < hi)
|
||||
{
|
||||
var mid = lo + (hi - lo) / 2;
|
||||
if (_ranges[mid].Start <= seq)
|
||||
lo = mid + 1;
|
||||
else
|
||||
hi = mid;
|
||||
}
|
||||
// lo == first index where Start > seq, so lo-1 is the last range with Start <= seq.
|
||||
var leftIdx = lo - 1; // may be -1 if no range has Start <= seq
|
||||
|
||||
// Check if seq is inside the range at leftIdx.
|
||||
if (leftIdx >= 0 && _ranges[leftIdx].End >= seq)
|
||||
return false; // already present
|
||||
|
||||
// At this point seq is not in any range. Determine insertion context.
|
||||
// leftIdx: the range immediately to the left of seq (Start <= seq, End < seq)
|
||||
// rightIdx = lo: the range immediately to the right of seq (Start > seq)
|
||||
var rightIdx = lo;
|
||||
|
||||
// Check adjacency with left neighbor (End + 1 == seq, i.e. seq extends the right edge).
|
||||
// Safe: if End == ulong.MaxValue the range would already contain seq.
|
||||
var leftAdjacent = leftIdx >= 0 && _ranges[leftIdx].End + 1 == seq;
|
||||
|
||||
// Check adjacency with right neighbor (Start - 1 == seq, i.e. seq extends the left edge).
|
||||
// Need to guard against Start == 0 (underflow), but if Start == 0 then Start > seq is impossible
|
||||
// since seq >= 0 and Start <= seq would have been caught by leftIdx check.
|
||||
var rightAdjacent = rightIdx < _ranges.Count && _ranges[rightIdx].Start > 0
|
||||
&& _ranges[rightIdx].Start - 1 == seq;
|
||||
|
||||
if (leftAdjacent && rightAdjacent)
|
||||
{
|
||||
// seq bridges the gap between the two neighbors → merge into one range.
|
||||
var newStart = _ranges[leftIdx].Start;
|
||||
var newEnd = _ranges[rightIdx].End;
|
||||
_ranges.RemoveAt(rightIdx); // remove right neighbor first (higher index)
|
||||
_ranges[leftIdx] = (newStart, newEnd);
|
||||
}
|
||||
else if (leftAdjacent)
|
||||
{
|
||||
// Extend the left neighbor's right edge to include seq.
|
||||
_ranges[leftIdx] = (_ranges[leftIdx].Start, seq);
|
||||
}
|
||||
else if (rightAdjacent)
|
||||
{
|
||||
// Extend the right neighbor's left edge to include seq.
|
||||
_ranges[rightIdx] = (seq, _ranges[rightIdx].End);
|
||||
}
|
||||
else
|
||||
{
|
||||
// No adjacency — insert a new single-element range at the correct position.
|
||||
_ranges.Insert(rightIdx, (seq, seq));
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Removes <paramref name="seq"/> from the set.
|
||||
/// Splits ranges if necessary.
|
||||
/// Returns <c>true</c> if the sequence was present.
|
||||
/// Reference: golang/nats-server/server/avl/seqset.go:80 (Delete).
|
||||
/// </summary>
|
||||
public bool Remove(ulong seq)
|
||||
{
|
||||
// Binary search for the range that contains seq.
|
||||
var lo = 0;
|
||||
var hi = _ranges.Count - 1;
|
||||
while (lo <= hi)
|
||||
{
|
||||
var mid = lo + (hi - lo) / 2;
|
||||
var (rs, re) = _ranges[mid];
|
||||
if (seq < rs)
|
||||
hi = mid - 1;
|
||||
else if (seq > re)
|
||||
lo = mid + 1;
|
||||
else
|
||||
{
|
||||
// Found the range [rs, re] that contains seq.
|
||||
if (rs == re)
|
||||
{
|
||||
// Single-element range → remove entirely.
|
||||
_ranges.RemoveAt(mid);
|
||||
}
|
||||
else if (seq == rs)
|
||||
{
|
||||
// Trim left edge.
|
||||
_ranges[mid] = (rs + 1, re);
|
||||
}
|
||||
else if (seq == re)
|
||||
{
|
||||
// Trim right edge.
|
||||
_ranges[mid] = (rs, re - 1);
|
||||
}
|
||||
else
|
||||
{
|
||||
// Split: [rs, seq-1] and [seq+1, re].
|
||||
_ranges[mid] = (seq + 1, re);
|
||||
_ranges.Insert(mid, (rs, seq - 1));
|
||||
}
|
||||
return true;
|
||||
}
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Returns <c>true</c> if <paramref name="seq"/> is a member of the set.
|
||||
/// Binary search: O(log R) where R is the number of distinct ranges.
|
||||
/// Reference: golang/nats-server/server/avl/seqset.go:52 (Exists).
|
||||
/// </summary>
|
||||
public bool Contains(ulong seq)
|
||||
{
|
||||
var lo = 0;
|
||||
var hi = _ranges.Count - 1;
|
||||
while (lo <= hi)
|
||||
{
|
||||
var mid = lo + (hi - lo) / 2;
|
||||
var (rs, re) = _ranges[mid];
|
||||
if (seq < rs)
|
||||
hi = mid - 1;
|
||||
else if (seq > re)
|
||||
lo = mid + 1;
|
||||
else
|
||||
return true;
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Removes all sequences from the set.
|
||||
/// Reference: golang/nats-server/server/avl/seqset.go:107 (Empty).
|
||||
/// </summary>
|
||||
public void Clear() => _ranges.Clear();
|
||||
|
||||
/// <summary>
|
||||
/// Copies all sequences in this set into a new <see cref="HashSet{T}"/>.
|
||||
/// Used when callers require an <see cref="IReadOnlySet{T}"/> snapshot.
|
||||
/// </summary>
|
||||
public HashSet<ulong> ToHashSet()
|
||||
{
|
||||
var set = new HashSet<ulong>(Count);
|
||||
foreach (var seq in this)
|
||||
set.Add(seq);
|
||||
return set;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Returns the number of distinct compressed ranges stored internally.
|
||||
/// </summary>
|
||||
internal int RangeCount => _ranges.Count;
|
||||
|
||||
/// <summary>
|
||||
/// Enumerates all sequences in ascending order.
|
||||
/// Reference: golang/nats-server/server/avl/seqset.go:122 (Range).
|
||||
/// </summary>
|
||||
public IEnumerator<ulong> GetEnumerator()
|
||||
{
|
||||
foreach (var (start, end) in _ranges)
|
||||
for (var seq = start; seq <= end; seq++)
|
||||
{
|
||||
yield return seq;
|
||||
if (seq == ulong.MaxValue) yield break; // prevent overflow
|
||||
}
|
||||
}
|
||||
|
||||
System.Collections.IEnumerator System.Collections.IEnumerable.GetEnumerator()
|
||||
=> GetEnumerator();
|
||||
}
|
||||
Reference in New Issue
Block a user