jie65535/JChatGPT
1
0
Fork 0
mirror of https://github.com/jie65535/JChatGPT.git synced 2026-09-15 02:56:10 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

profile: optimize conversation analysis concurrency

Browse Source
This commit is contained in:
jie65535
2026-08-03 16:39:19 +08:00
parent b96b732b92
commit 94f303ec72
12 changed files with 328 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/main/kotlin/command/PluginCommands.kt
+8 -3
View File
@@ -20,6 +20,7 @@ import top.jie65535.mirai.data.PluginData
import top.jie65535.mirai.data.SkillStore
import top.jie65535.mirai.data.TokenUsageStore
import top.jie65535.mirai.llm.LargeLanguageModels
import top.jie65535.mirai.llm.normalizeMaxConcurrentRequests
import top.jie65535.mirai.profile.GroupProfileAnalysisReport
import top.jie65535.mirai.profile.ProfileAnalysisReport
import top.jie65535.mirai.profile.ProfileAutoMaintenance
@@ -90,18 +91,22 @@ object PluginCommands : CompositeCommand(
require(batches > 0) { "batches 必须是正数" }
val analyzeAllGroups = groupIds.isBlank()
val parsedGroupIds = if (analyzeAllGroups) {
UserProfileAnalysisService.listHistoryGroupIds()
UserProfileAnalysisService.listPendingHistoryGroupIds()
} else {
parseProfileGroupIds(groupIds)
}
if (parsedGroupIds.isEmpty()) {
sendMessage("聊天记录中没有可分析的群消息。")
sendMessage("没有待推进的群画像：历史库中无有效群消息，或所有群均已追平当前快照。")
return
}
val runToken = UserProfileAnalysisService.newRunToken()
sendMessage(
"已启动 ${parsedGroupIds.size} 个群的画像分析，每群最多推进 $batches 个批次。" +
if (analyzeAllGroups) " 全量模式不设并发上限。" else ""
if (analyzeAllGroups) {
" 画像请求并发上限 ${normalizeMaxConcurrentRequests(PluginConfig.profileMaxConcurrentRequests)}。"
} else {
""
}
)
val completedGroups = AtomicInteger()
val successfulGroups = AtomicInteger()
src/main/kotlin/config/PluginConfig.kt
+3
View File
@@ -75,6 +75,9 @@ object PluginConfig : AutoSavePluginConfig("Config") {
@ValueDescription("画像分析模型额外请求体JSON。留空时继承聊天模型额外请求体")
val profileModelExtraBody: String by value("")
@ValueDescription("画像模型同时执行的请求上限，取值1~512，默认128；应用层排队且不计入首块超时，仅影响画像模型")
val profileMaxConcurrentRequests: Int by value(128)
@ValueDescription("画像分析使用的聊天记录SQLite路径。留空时使用插件自己的chat-history.sqlite；本地实验可填写历史库绝对路径")
val profileHistoryDatabasePath: String by value("")
src/main/kotlin/llm/LargeLanguageModels.kt
+1
View File
@@ -171,6 +171,7 @@ object LargeLanguageModels {
timeout = maxOf(timeout, profileFirstChunk),
firstChunkTimeout = profileFirstChunk,
extraBody = parseExtraBody(extraBody),
maxConcurrentRequests = PluginConfig.profileMaxConcurrentRequests,
),
model = model,
temperature = PluginConfig.profileModelTemperature ?: PluginConfig.chatTemperature,
src/main/kotlin/llm/ModelService.kt
+39 -2
View File
@@ -12,10 +12,14 @@ import io.ktor.http.*
import io.ktor.utils.io.*
import kotlinx.coroutines.currentCoroutineContext
import kotlinx.coroutines.flow.Flow
import kotlinx.coroutines.flow.collect
import kotlinx.coroutines.flow.flow
import kotlinx.coroutines.isActive
import kotlinx.coroutines.sync.Semaphore
import kotlinx.coroutines.sync.withPermit
import kotlinx.coroutines.withTimeout
import kotlinx.serialization.json.*
import okhttp3.Dispatcher as OkHttpDispatcher
import kotlin.time.Duration
class ModelService(
@@ -23,10 +27,21 @@ class ModelService(
val token: String,
val timeout: Duration,
val firstChunkTimeout: Duration,
val extraBody: JsonObject? = null
val extraBody: JsonObject? = null,
maxConcurrentRequests: Int? = null,
) {
private val maxConcurrentRequests = maxConcurrentRequests?.let(::normalizeMaxConcurrentRequests)
private val requestSemaphore = this.maxConcurrentRequests?.let(::Semaphore)
val httpClient: HttpClient by lazy {
HttpClient(OkHttp) {
this@ModelService.maxConcurrentRequests?.let { concurrencyLimit ->
engine {
config {
dispatcher(createRequestDispatcher(concurrencyLimit))
}
}
}
install(HttpTimeout) {
// 流式响应的「首 token」与「token 间隔」超时统一由应用层 withTimeout 管控（见 chatCompletions）。
// 这里特意不设 requestTimeoutMillis：否则正常但耗时较长的流式输出会被 Ktor 在中途整体掐断。
@@ -79,7 +94,7 @@ class ModelService(
}
val body = JsonObject(requestJson).toString()
return flow {
val responseFlow: Flow<ChatCompletionChunk> = flow {
// 关键：服务器繁忙时会拖住「响应头」，使 httpClient.post() 自身阻塞在等待响应的阶段，
// 因此必须把 post() 连同首个 data 块的读取一起包进 withTimeout。
// 否则首 token 超时永远不会触发（post() 还没返回，根本进不到读取循环），
@@ -137,5 +152,27 @@ class ModelService(
channel?.cancel()
}
}
return responseFlow.withConcurrencyLimit(requestSemaphore)
}
}
internal const val MAX_MODEL_CONCURRENT_REQUESTS = 512
internal fun normalizeMaxConcurrentRequests(value: Int): Int =
value.coerceIn(1, MAX_MODEL_CONCURRENT_REQUESTS)
internal fun createRequestDispatcher(maxConcurrentRequests: Int): OkHttpDispatcher =
OkHttpDispatcher().apply {
val concurrencyLimit = normalizeMaxConcurrentRequests(maxConcurrentRequests)
maxRequests = concurrencyLimit
maxRequestsPerHost = concurrencyLimit
}
internal fun <T> Flow<T>.withConcurrencyLimit(semaphore: Semaphore?): Flow<T> {
semaphore ?: return this
return flow {
semaphore.withPermit {
this@withConcurrencyLimit.collect { value -> emit(value) }
}
}
}
src/main/kotlin/profile/ProfileHistoryReader.kt
+24 -6
View File
@@ -13,7 +13,12 @@ import java.sql.ResultSet
class ProfileHistoryReader(private val databaseFile: File) {
data class TimeBounds(val startTime: Int, val endTime: Int)
data class GroupTimeBounds(val botId: Long, val startTime: Int, val endTime: Int)
data class GroupTimeBounds(
val botId: Long,
val groupId: Long,
val startTime: Int,
val endTime: Int,
)
private data class Episode(
val index: Int,
@@ -63,6 +68,7 @@ class ProfileHistoryReader(private val databaseFile: File) {
if (!results.next()) return@use null
GroupTimeBounds(
botId = results.getLong("bot_id"),
groupId = groupId,
startTime = results.getInt("min_time"),
endTime = results.getInt("max_time").safeNextSecond(),
)
@@ -70,20 +76,32 @@ class ProfileHistoryReader(private val databaseFile: File) {
}
}
fun listGroupIds(): List<Long> = openReadConnection().use { connection ->
fun listGroupTimeBounds(): List<GroupTimeBounds> = openReadConnection().use { connection ->
connection.prepareStatement(
"""
SELECT target_id, MAX(time) AS last_message_time
SELECT bot_id, target_id, MIN(time) AS min_time, MAX(time) AS max_time
FROM message_record
WHERE kind = ? AND recalled = 0 AND target_id > 0
GROUP BY target_id
ORDER BY last_message_time DESC, target_id ASC
GROUP BY bot_id, target_id
ORDER BY max_time DESC, target_id ASC, bot_id ASC
""".trimIndent()
).use { statement ->
statement.setInt(1, MessageSourceKind.GROUP.ordinal)
statement.executeQuery().use { results ->
val seenGroupIds = hashSetOf<Long>()
buildList {
while (results.next()) add(results.getLong("target_id"))
while (results.next()) {
val groupId = results.getLong("target_id")
if (!seenGroupIds.add(groupId)) continue
add(
GroupTimeBounds(
botId = results.getLong("bot_id"),
groupId = groupId,
startTime = results.getInt("min_time"),
endTime = results.getInt("max_time").safeNextSecond(),
)
)
}
}
}
}
src/main/kotlin/profile/UserProfileAnalysisService.kt
+102 -34
View File
@@ -14,6 +14,8 @@ import java.security.MessageDigest
import java.util.concurrent.ConcurrentHashMap
object UserProfileAnalysisService {
private const val MAX_CONVERSATION_CONFLICT_RETRIES = 3
private val runningUsers = ConcurrentHashMap.newKeySet<Long>()
private val runningGroups = ConcurrentHashMap.newKeySet<Long>()
private val runningCompactions = ConcurrentHashMap.newKeySet<Long>()
@@ -32,11 +34,19 @@ object UserProfileAnalysisService {
return report
}
suspend fun listHistoryGroupIds(): List<Long> {
suspend fun listPendingHistoryGroupIds(): List<Long> {
check(PluginConfig.profileEnabled) { "历史用户画像分析未启用" }
check(UserProfileStore.isAvailable) { "用户画像数据库不可用" }
return withContext(Dispatchers.IO) {
ProfileHistoryReader(resolveHistoryFile()).listGroupIds()
val historyBounds = ProfileHistoryReader(resolveHistoryFile()).listGroupTimeBounds()
val cursors = UserProfileStore.loadGroupCursors()
.associateBy { cursor -> cursor.botId to cursor.groupId }
historyBounds.asSequence()
.filter { bounds ->
isGroupAnalysisPending(bounds, cursors[bounds.botId to bounds.groupId])
}
.map(ProfileHistoryReader.GroupTimeBounds::groupId)
.toList()
}
}
@@ -274,10 +284,6 @@ object UserProfileAnalysisService {
val reader = withContext(Dispatchers.IO) { ProfileHistoryReader(resolveHistoryFile()) }
val bounds = withContext(Dispatchers.IO) { reader.findGroupTimeBounds(groupId) }
?: return emptyGroupReport(groupId)
val endpoint = checkNotNull(LargeLanguageModels.profile) {
"画像分析模型未配置，请设置 profileModelApi/profileModelToken，或配置可继承的聊天模型接入点"
}
val model: ConversationProfileModel = ProfileModelClient(endpoint)
var cursor = withContext(Dispatchers.IO) {
UserProfileStore.loadGroupCursor(bounds.botId, groupId)
} ?: GroupProfileCursor(
@@ -297,6 +303,12 @@ object UserProfileAnalysisService {
var skippedOperations = 0
var totalUsage = ProfileTokenUsage()
var caughtUp = cursor.cursorTime >= cursor.snapshotEndTime
val model: ConversationProfileModel by lazy {
val endpoint = checkNotNull(LargeLanguageModels.profile) {
"画像分析模型未配置，请设置 profileModelApi/profileModelToken，或配置可继承的聊天模型接入点"
}
ProfileModelClient(endpoint)
}
while (processedBatches < maxBatches && !caughtUp && runGate.canContinue(runToken)) {
val batch = withContext(Dispatchers.IO) {
@@ -421,20 +433,16 @@ object UserProfileAnalysisService {
.filterValues { it >= minAuthoredTextChars.coerceAtLeast(1) }
.keys
if (eligibleUserIds.isEmpty()) return null
return userLocks.withUserLocks(eligibleUserIds) locked@{
if (withContext(Dispatchers.IO) { UserProfileStore.isConversationProcessed(batch.inputHash) }) {
return@locked null
}
val profiles = withContext(Dispatchers.IO) {
eligibleUserIds.associateWith { userId ->
UserProfileStore.load(userId) ?: UserProfileSnapshot(
userId = userId,
cursorTime = 0,
snapshotEndTime = 0,
)
}
var profiles = userLocks.withUserLocks(eligibleUserIds) {
withContext(Dispatchers.IO) {
if (UserProfileStore.isConversationProcessed(batch.inputHash)) null
else loadConversationProfiles(eligibleUserIds)
}
} ?: return null
var conflictRetries = 0
var totalUsage = ProfileTokenUsage()
while (true) {
val (result, reductions) = analyzeConversationWithRetry(
model = model,
profiles = profiles,
@@ -444,25 +452,72 @@ object UserProfileAnalysisService {
summaryMaxLength = summaryMaxLength,
onRetryFailure = onRetryFailure,
)
withContext(Dispatchers.IO) {
UserProfileStore.commitConversation(
reductions = reductions.map { reduction -> reduction to batch.forUser(reduction.profile.userId) },
usage = result.usage,
)
totalUsage += result.usage
val commitOutcome = userLocks.withUserLocks(eligibleUserIds) {
withContext(Dispatchers.IO) {
if (UserProfileStore.isConversationProcessed(batch.inputHash)) {
ConversationCommitOutcome.AlreadyProcessed
} else {
val latestProfiles = loadConversationProfiles(eligibleUserIds)
if (hasProfileVersionConflict(profiles, latestProfiles)) {
ConversationCommitOutcome.Conflict(latestProfiles)
} else {
UserProfileStore.commitConversation(
reductions = reductions.map { reduction ->
reduction to batch.forUser(reduction.profile.userId)
},
usage = totalUsage,
)
ConversationCommitOutcome.Committed
}
}
}
}
onCommittedOperations(
"source=CONVERSATION bot=${batch.botId} group=${batch.groupId} " +
"batch=[${batch.startTime},${batch.endTime})",
reductions,
)
ConversationProfileAnalysisReport(
analyzedUsers = eligibleUserIds.size,
processedMessages = batch.messages.size,
appliedOperations = reductions.sumOf { it.operations.size },
skippedOperations = reductions.sumOf { it.skippedOperations.size },
usage = result.usage,
when (commitOutcome) {
ConversationCommitOutcome.AlreadyProcessed -> return null
ConversationCommitOutcome.Committed -> {
onCommittedOperations(
"source=CONVERSATION bot=${batch.botId} group=${batch.groupId} " +
"batch=[${batch.startTime},${batch.endTime})",
reductions,
)
return ConversationProfileAnalysisReport(
analyzedUsers = eligibleUserIds.size,
processedMessages = batch.messages.size,
appliedOperations = reductions.sumOf { it.operations.size },
skippedOperations = reductions.sumOf { it.skippedOperations.size },
usage = totalUsage,
)
}
is ConversationCommitOutcome.Conflict -> {
conflictRetries++
if (conflictRetries > MAX_CONVERSATION_CONFLICT_RETRIES) {
throw IllegalStateException(
"群 ${batch.groupId} 会话画像提交连续冲突 $conflictRetries 次，未提交结果"
)
}
profiles = commitOutcome.latestProfiles
}
}
}
}
private fun loadConversationProfiles(userIds: Set<Long>): Map<Long, UserProfileSnapshot> =
userIds.associateWith { userId ->
UserProfileStore.load(userId) ?: UserProfileSnapshot(
userId = userId,
cursorTime = 0,
snapshotEndTime = 0,
)
}
private fun hasProfileVersionConflict(
expectedProfiles: Map<Long, UserProfileSnapshot>,
latestProfiles: Map<Long, UserProfileSnapshot>,
): Boolean = expectedProfiles.any { (userId, expected) ->
latestProfiles[userId]?.version != expected.version
}
private suspend fun analyzeConversationWithRetry(
@@ -698,4 +753,17 @@ object UserProfileAnalysisService {
if (skipped.size > 8) "；其余 ${skipped.size - 8} 项已省略" else ""
)
}
private sealed class ConversationCommitOutcome {
object Committed : ConversationCommitOutcome()
object AlreadyProcessed : ConversationCommitOutcome()
data class Conflict(
val latestProfiles: Map<Long, UserProfileSnapshot>,
) : ConversationCommitOutcome()
}
}
internal fun isGroupAnalysisPending(
bounds: ProfileHistoryReader.GroupTimeBounds,
cursor: GroupProfileCursor?,
): Boolean = cursor == null || cursor.cursorTime < maxOf(cursor.snapshotEndTime, bounds.endTime)
src/main/kotlin/profile/UserProfileStore.kt
+28 -7
View File
@@ -192,13 +192,26 @@ object UserProfileStore {
statement.setLong(2, groupId)
statement.executeQuery().use { results ->
if (!results.next()) return@use null
GroupProfileCursor(
botId = results.getLong("bot_id"),
groupId = results.getLong("group_id"),
cursorTime = results.getInt("cursor_time"),
snapshotEndTime = results.getInt("snapshot_end_time"),
updatedAt = results.getLong("updated_at"),
)
results.toGroupProfileCursor()
}
}
}
}
fun loadGroupCursors(): List<GroupProfileCursor> {
check(initialized) { "用户画像数据库尚未初始化" }
return openReadConnection().use { connection ->
connection.prepareStatement(
"""
SELECT bot_id, group_id, cursor_time, snapshot_end_time, updated_at
FROM profile_group_cursor
ORDER BY bot_id, group_id
""".trimIndent()
).use { statement ->
statement.executeQuery().use { results ->
buildList {
while (results.next()) add(results.toGroupProfileCursor())
}
}
}
}
@@ -602,6 +615,14 @@ object UserProfileStore {
lastConfirmedAt = getInt("last_confirmed_at"),
)
private fun ResultSet.toGroupProfileCursor() = GroupProfileCursor(
botId = getLong("bot_id"),
groupId = getLong("group_id"),
cursorTime = getInt("cursor_time"),
snapshotEndTime = getInt("snapshot_end_time"),
updatedAt = getLong("updated_at"),
)
private fun ProfileCategory.toStorageValue(): String = name.lowercase()
private fun ProfileConfidence.toStorageValue(): String = name.lowercase()
private fun Int.safeNextSecond(): Int = if (this == Int.MAX_VALUE) this else this + 1