This commit is contained in:
Chris Rickman
2026-03-27 15:39:04 -07:00
Unverified
parent 7d6beea6e5
commit 64e529916c
4 changed files with 65 additions and 53 deletions
@@ -17,21 +17,18 @@ public enum CompactionApproach
/// <summary>
/// Applies the lightest available compaction techniques.
/// Collapses old tool call groups into concise summaries and uses truncation as an emergency backstop.
/// Does not require a summarization <see cref="Microsoft.Extensions.AI.IChatClient"/>.
/// </summary>
Gentle,
/// <summary>
/// Balances context preservation with compaction efficiency.
/// Applies tool result collapsing, LLM-based summarization, and truncation as an emergency backstop.
/// Requires a summarization <see cref="Microsoft.Extensions.AI.IChatClient"/>.
/// </summary>
Balanced,
/// <summary>
/// Applies the most aggressive available compaction techniques.
/// Applies tool result collapsing, LLM-based summarization, turn-based sliding window, and truncation.
/// Requires a summarization <see cref="Microsoft.Extensions.AI.IChatClient"/>.
/// </summary>
Aggressive,
}
@@ -2,7 +2,6 @@
using System;
using Microsoft.Extensions.AI;
using Microsoft.Shared.Diagnostics;
namespace Microsoft.Agents.AI.Compaction;
@@ -62,16 +61,12 @@ public abstract partial class CompactionStrategy
/// </remarks>
/// <param name="approach">
/// The compaction approach that controls which strategy or pipeline to use.
/// <see cref="CompactionApproach.Gentle"/> does not require a <paramref name="chatClient"/>;
/// <see cref="CompactionApproach.Balanced"/> and <see cref="CompactionApproach.Aggressive"/> require one.
/// </param>
/// <param name="size">
/// The context-size profile that controls token and message thresholds.
/// </param>
/// <param name="chatClient">
/// The <see cref="IChatClient"/> used for LLM-based summarization.
/// Required when <paramref name="approach"/> is <see cref="CompactionApproach.Balanced"/> or
/// <see cref="CompactionApproach.Aggressive"/>; ignored for <see cref="CompactionApproach.Gentle"/>.
/// </param>
/// <returns>A <see cref="CompactionStrategy"/> configured for the specified approach and size.</returns>
/// <exception cref="ArgumentNullException">
@@ -80,21 +75,16 @@ public abstract partial class CompactionStrategy
/// <exception cref="ArgumentOutOfRangeException">
/// <paramref name="approach"/> or <paramref name="size"/> is not a defined enum value.
/// </exception>
public static CompactionStrategy Create(CompactionApproach approach, CompactionSize size, IChatClient? chatClient = null)
public static CompactionStrategy Create(CompactionApproach approach, CompactionSize size, IChatClient chatClient)
{
if (approach is CompactionApproach.Balanced or CompactionApproach.Aggressive)
{
_ = Throw.IfNull(chatClient);
}
int tokenLimit = GetTokenLimit(size);
int messageLimit = GetMessageLimit(size);
return approach switch
{
CompactionApproach.Gentle => CreateGentlePipeline(tokenLimit, messageLimit),
CompactionApproach.Balanced => CreateBalancedPipeline(tokenLimit, messageLimit, chatClient!),
CompactionApproach.Aggressive => CreateAggressivePipeline(tokenLimit, messageLimit, GetTurnLimit(size), chatClient!),
CompactionApproach.Gentle => CreateGentlePipeline(tokenLimit, messageLimit, chatClient),
CompactionApproach.Balanced => CreateBalancedPipeline(tokenLimit, messageLimit, chatClient),
CompactionApproach.Aggressive => CreateAggressivePipeline(tokenLimit, messageLimit, GetTurnLimit(size), chatClient),
_ => throw new ArgumentOutOfRangeException(nameof(approach), approach, null),
};
}
@@ -123,35 +113,59 @@ public abstract partial class CompactionStrategy
_ => throw new ArgumentOutOfRangeException(nameof(size), size, null),
};
private static PipelineCompactionStrategy CreateGentlePipeline(int tokenLimit, int messageLimit) =>
new(
new ToolResultCompactionStrategy(CompactionTriggers.MessagesExceed(messageLimit)),
new TruncationCompactionStrategy(CompactionTriggers.TokensExceed(tokenLimit)));
private static PipelineCompactionStrategy CreateGentlePipeline(int tokenLimit, int messageLimit, IChatClient chatClient)
{
int messageTarget = messageLimit * 4 / 5;
int tokenTarget = tokenLimit * 4 / 5;
return new(
new ToolResultCompactionStrategy(
trigger: CompactionTriggers.MessagesExceed(messageLimit),
target: CompactionTriggers.MessagesBelow(messageTarget)),
new SummarizationCompactionStrategy(
chatClient,
trigger: CompactionTriggers.TokensExceed(tokenLimit),
target: CompactionTriggers.TokensBelow(tokenTarget)));
}
private static PipelineCompactionStrategy CreateBalancedPipeline(int tokenLimit, int messageLimit, IChatClient chatClient)
{
// Early stages trigger at two-thirds of the limit so the pipeline has room to compact
// incrementally before reaching the emergency truncation backstop at the full limit.
int earlyMessageTrigger = messageLimit * 2 / 3;
int earlyTokenTrigger = tokenLimit * 2 / 3;
int messageTarget = messageLimit * 3 / 4;
int tokenTarget = tokenLimit * 4 / 5;
return new(
new ToolResultCompactionStrategy(CompactionTriggers.MessagesExceed(earlyMessageTrigger)),
new SummarizationCompactionStrategy(chatClient, CompactionTriggers.TokensExceed(earlyTokenTrigger)),
new TruncationCompactionStrategy(CompactionTriggers.TokensExceed(tokenLimit)));
new ToolResultCompactionStrategy(
trigger: CompactionTriggers.MessagesExceed(messageLimit),
target: CompactionTriggers.MessagesBelow(messageTarget)),
new SummarizationCompactionStrategy(
chatClient,
trigger: CompactionTriggers.TokensExceed(tokenLimit),
target: CompactionTriggers.TokensBelow(tokenTarget)),
new TruncationCompactionStrategy(
trigger: CompactionTriggers.TokensExceed(tokenLimit),
target: CompactionTriggers.TokensBelow(tokenTarget)));
}
private static PipelineCompactionStrategy CreateAggressivePipeline(int tokenLimit, int messageLimit, int turnLimit, IChatClient chatClient)
{
// Early stages trigger at half the limit so compaction kicks in sooner and
// the sliding window and truncation backstop are reached less often.
int earlyMessageTrigger = messageLimit / 2;
int earlyTokenTrigger = tokenLimit / 2;
int messageTarget = messageLimit * 3 / 4;
int tokenTarget = tokenLimit * 3 / 4;
return new(
new ToolResultCompactionStrategy(CompactionTriggers.MessagesExceed(earlyMessageTrigger)),
new SummarizationCompactionStrategy(chatClient, CompactionTriggers.TokensExceed(earlyTokenTrigger)),
new SlidingWindowCompactionStrategy(CompactionTriggers.TurnsExceed(turnLimit)),
new TruncationCompactionStrategy(CompactionTriggers.TokensExceed(tokenLimit)));
new ToolResultCompactionStrategy(
trigger: CompactionTriggers.MessagesExceed(messageLimit),
target: CompactionTriggers.MessagesBelow(messageTarget)),
new SummarizationCompactionStrategy(
chatClient,
trigger: CompactionTriggers.TokensExceed(tokenLimit),
target: CompactionTriggers.TokensBelow(tokenTarget)),
new SlidingWindowCompactionStrategy(
trigger: CompactionTriggers.TokensExceed(tokenLimit),
target: CompactionTriggers.TokensBelow(tokenTarget)),
new TruncationCompactionStrategy(
trigger: CompactionTriggers.TokensExceed(tokenLimit),
target: CompactionTriggers.TokensBelow(tokenTarget)));
}
}
@@ -50,6 +50,14 @@ public static class CompactionTriggers
public static CompactionTrigger TokensExceed(int maxTokens) =>
index => index.IncludedTokenCount > maxTokens;
/// <summary>
/// Creates a trigger that fires when the included message count is below the specified maximum.
/// </summary>
/// <param name="maxMessages">The message threshold.</param>
/// <returns>A <see cref="CompactionTrigger"/> that evaluates included message count.</returns>
public static CompactionTrigger MessagesBelow(int maxMessages) =>
index => index.IncludedMessageCount < maxMessages;
/// <summary>
/// Creates a trigger that fires when the included message count exceeds the specified maximum.
/// </summary>
@@ -33,11 +33,11 @@ public class CompactionStrategyCreateTests
{
PipelineCompactionStrategy pipeline =
Assert.IsType<PipelineCompactionStrategy>(
CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Compact));
CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Compact, CreateMockChatClient()));
Assert.Equal(2, pipeline.Strategies.Count);
Assert.IsType<ToolResultCompactionStrategy>(pipeline.Strategies[0]);
Assert.IsType<TruncationCompactionStrategy>(pipeline.Strategies[1]);
Assert.IsType<SummarizationCompactionStrategy>(pipeline.Strategies[1]);
}
[Fact]
@@ -45,11 +45,11 @@ public class CompactionStrategyCreateTests
{
PipelineCompactionStrategy pipeline =
Assert.IsType<PipelineCompactionStrategy>(
CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Moderate));
CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Moderate, CreateMockChatClient()));
Assert.Equal(2, pipeline.Strategies.Count);
Assert.IsType<ToolResultCompactionStrategy>(pipeline.Strategies[0]);
Assert.IsType<TruncationCompactionStrategy>(pipeline.Strategies[1]);
Assert.IsType<SummarizationCompactionStrategy>(pipeline.Strategies[1]);
}
[Fact]
@@ -57,18 +57,18 @@ public class CompactionStrategyCreateTests
{
PipelineCompactionStrategy pipeline =
Assert.IsType<PipelineCompactionStrategy>(
CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Generous));
CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Generous, CreateMockChatClient()));
Assert.Equal(2, pipeline.Strategies.Count);
Assert.IsType<ToolResultCompactionStrategy>(pipeline.Strategies[0]);
Assert.IsType<TruncationCompactionStrategy>(pipeline.Strategies[1]);
Assert.IsType<SummarizationCompactionStrategy>(pipeline.Strategies[1]);
}
[Fact]
public void CreateGentleDoesNotRequireChatClient()
{
// No chatClient supplied — should succeed without throwing.
CompactionStrategy strategy = CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Moderate);
CompactionStrategy strategy = CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Moderate, CreateMockChatClient());
Assert.NotNull(strategy);
}
@@ -117,7 +117,7 @@ public class CompactionStrategyCreateTests
public void CreateBalancedNullChatClientThrows()
{
Assert.Throws<ArgumentNullException>(
() => CompactionStrategy.Create(CompactionApproach.Balanced, CompactionSize.Moderate, null));
() => CompactionStrategy.Create(CompactionApproach.Balanced, CompactionSize.Moderate, null!));
}
// ── Aggressive ────────────────────────────────────────────────────────────
@@ -164,27 +164,20 @@ public class CompactionStrategyCreateTests
Assert.IsType<TruncationCompactionStrategy>(pipeline.Strategies[3]);
}
[Fact]
public void CreateAggressiveNullChatClientThrows()
{
Assert.Throws<ArgumentNullException>(
() => CompactionStrategy.Create(CompactionApproach.Aggressive, CompactionSize.Moderate, null));
}
// ── Invalid enum values ───────────────────────────────────────────────────
[Fact]
public void CreateInvalidApproachThrows()
{
Assert.Throws<ArgumentOutOfRangeException>(
() => CompactionStrategy.Create((CompactionApproach)99, CompactionSize.Moderate));
() => CompactionStrategy.Create((CompactionApproach)99, CompactionSize.Moderate, CreateMockChatClient()));
}
[Fact]
public void CreateInvalidSizeThrows()
{
Assert.Throws<ArgumentOutOfRangeException>(
() => CompactionStrategy.Create(CompactionApproach.Gentle, (CompactionSize)99));
() => CompactionStrategy.Create(CompactionApproach.Gentle, (CompactionSize)99, CreateMockChatClient()));
}
// ── Size-threshold behavioral verification ────────────────────────────────
@@ -203,8 +196,8 @@ public class CompactionStrategyCreateTests
// The configuration preserves a number of most-recent groups, leaving the oldest
// tool-call group eligible for collapsing, which makes the behavioral difference
// between Compact and Moderate sizes observable in this test.
CompactionStrategy compactPipeline = CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Compact);
CompactionStrategy moderatePipeline = CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Moderate);
CompactionStrategy compactPipeline = CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Compact, CreateMockChatClient());
CompactionStrategy moderatePipeline = CompactionStrategy.Create(CompactionApproach.Gentle, CompactionSize.Moderate, CreateMockChatClient());
List<ChatMessage> messages =
[