perf(session): deduplicate Redis session response bodies (#1415) - #1417
Conversation
Replace three independent SETEX writes for legacy, before, and after response bodies with a single atomic storeSessionResponseBodySet call that writes one request-scoped Redis Hash. Identical views share a single body field via intra-Hash refs, reducing Redis memory from 3x the response size to 1x for the common case where all three views are identical. The new layout uses Lua scripts to atomically replace the prior generation (Hash plus legacy keys) on every write, preventing stale bodies from mixing across retries or concurrent writers. An authoritative over-budget marker prevents fallback to stale legacy keys when the aggregate unique-body byte count exceeds SESSION_RESPONSE_BODY_MAX_BYTES. A reader-first two-phase rollout is enforced via SESSION_RESPONSE_BODY_DEDUP_ENABLED (default false). Phase A deploys new readers while the flag stays false, writing legacy keys plus a layout=legacy marker. Phase B flips the flag to switch writers to the dedup Hash layout. Readers transparently handle both layouts and fall back to legacy keys only when no bundle Hash exists. Load fixture harness extended with exact-byte SSE sizing, complete response mode, BGSAVE triggering, and a Redis inspection script that validates bundle invariants, TTL cleanup, and container OOM state. Closes #1415
…eration switches The read Lua script previously declared only the dedup Hash key. When a writer switched generations between the Hash read and a subsequent legacy-key GET, the reader could observe a stale legacy marker pointing at keys already deleted by the next generation, returning null instead of the stored body. The script now declares both the Hash key and the target view's legacy key in a single Redis command, reading layout/present/ref/body and the legacy body atomically. This eliminates the separate legacy fallback GET entirely, removing the legacyFallback flag from the read result type and simplifying both call sites in getSessionResponse and getSessionDetailSnapshot. The detail-snapshot test mock now simulates the legacy key store so the atomic read returns the correct body. The dedup test adds a scenario where a writer clears legacy keys and installs a dedup Hash after the read fires, verifying the reader still returns the original legacy generation.
📝 WalkthroughWalkthrough新增响应正文 bundle 去重存储。系统支持三种正文视图、旧键兼容、Lua 原子读写、UTF-8 聚合字节限制和 TTL 管理。代理路径、配置、测试及 5 MiB 负载验收工具同步更新。 Changes会话响应正文去重
Estimated code review effort: 4 (Complex) | ~60 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches 💡 2📝 Generate docstrings 💡
🛠️ Fix failing CI checks 💡
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| function buildSessionResponseBodyBundleKey(sessionId: string, sequence: number): string { | ||
| return `session:${sessionId}:req:${sequence}:response-bodies:v1`; | ||
| } |
There was a problem hiding this comment.
The new response-bodies:v1 keys are not included in terminateSession, so terminating a session leaves potentially multi-megabyte response bodies in Redis until their normal TTL expires, reducing the immediacy and memory effectiveness of administrative cleanup.
Knowledge Base Used: Redis Caching and Session Tracking
Prompt To Fix With AI
This is a comment left during a code review.
Path: src/lib/session-manager.ts
Line: 186-188
Comment:
**Bundle keys bypass session cleanup**
The new `response-bodies:v1` keys are not included in `terminateSession`, so terminating a session leaves potentially multi-megabyte response bodies in Redis until their normal TTL expires, reducing the immediacy and memory effectiveness of administrative cleanup.
**Knowledge Base Used:** [Redis Caching and Session Tracking](https://app.greptile.com/ygxz/-/custom-context/knowledge-base/ding113/claude-code-hub/-/docs/redis-caching-and-sessions.md)
---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (3)
docs/research/issue-1415-session-response-body-dedup.md (1)
46-47: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value缺少空格。
writer省略之间缺少空格，与全文的中英混排风格不一致。📝 建议修改
-如果新 writer 把旧 key 改写为引用描述符, 旧 reader 会把描述符当正文; 如果新 writer省略重复 +如果新 writer 把旧 key 改写为引用描述符, 旧 reader 会把描述符当正文; 如果新 writer 省略重复 key, 旧 reader 会返回 `null`.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/research/issue-1415-session-response-body-dedup.md` around lines 46 - 47, 在文档描述中，将“writer省略”修改为“writer 省略”，补充中英文之间缺少的空格，并保持其余内容不变。tests/load/issue-1408-replay-oom/mock-upstream.cjs (1)
186-193: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick win每个请求都重新构建 5 MiB 帧数组。
buildSseFrames(responseBytes, responseMode)的输入在整个进程生命周期内是常量。当前每个请求都重新分配约 5 MiB 的字符串数组，会增加 mock 自身的 CPU 与内存压力，并干扰负载观测。请在模块加载时构建一次并复用。♻️ 建议重构
+const staticFrames = buildSseFrames(responseBytes, responseMode); + // 在请求处理中： - const frames = buildSseFrames(responseBytes, responseMode); + const frames = staticFrames;🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tests/load/issue-1408-replay-oom/mock-upstream.cjs` around lines 186 - 193, 将 buildSseFrames(responseBytes, responseMode) 从每次请求的处理流程移到模块加载阶段，仅构建一次并缓存结果；在 writeNext 及其所属请求逻辑中复用该模块级 frames，保持现有发送顺序与计数行为不变。tests/load/issue-1408-replay-oom/run-wave.sh (1)
100-118: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueTTL 轮询会隐藏所有失败原因。
2>/dev/null会丢弃inspect-redis.cjs的全部 stderr。如果失败原因不是 TTL 未过期（例如 docker 不可用或 manifest 解析失败），脚本会静默重试到超时。建议把 stderr 写入临时文件，并在超时时一并输出。同时建议校验
CCH_TTL_CLEANUP_TIMEOUT_SECONDS与CCH_RDB_DELAY_SECONDS为整数，避免[ ... -ge ]与sleep报错。🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tests/load/issue-1408-replay-oom/run-wave.sh` around lines 100 - 118, 更新 TTL 轮询中 inspect-redis.cjs 的错误处理，将 stderr 重定向到独立临时文件而不是丢弃，并在超时时输出该文件内容。为 CCH_TTL_CLEANUP_TIMEOUT_SECONDS 和 CCH_RDB_DELAY_SECONDS 增加整数格式校验，在进入超时计算或 sleep 前拒绝无效值并返回清晰错误。
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.env.example:
- Around line 108-113: Update the comments for
SESSION_RESPONSE_BODY_DEDUP_ENABLED to limit the atomic generation-marker
coordination guarantee to requests with a valid requestSequence. Explicitly note
that requests without a sequence use the compatibility legacy/before/after key
path and do not execute generation-marker Lua, while preserving the existing
deployment guidance.
In `@docs/research/issue-1415-session-response-body-dedup.md`:
- Around line 37-39: 修改文档中以 `#1408` 和 `#1415` 开头的行，避免将 issue 引用放在 Markdown
行首；将引用移入行内或改写为不触发标题解析的文本，同时保留原有内容和 issue 链接含义。
In `@tests/load/issue-1408-replay-oom/drive-disconnect-waves.cjs`:
- Around line 184-192: 为所有 handle 的 completion Promise 附加 no-op catch，不要仅在
requestMode 为 disconnect 时处理；同时更新 waitForCompletions 使用已附加 rejection 处理的 Promise
副本，确保 complete 模式及超时后 abortRequests 导致的后续 rejection 不会未处理。
---
Nitpick comments:
In `@docs/research/issue-1415-session-response-body-dedup.md`:
- Around line 46-47: 在文档描述中，将“writer省略”修改为“writer 省略”，补充中英文之间缺少的空格，并保持其余内容不变。
In `@tests/load/issue-1408-replay-oom/mock-upstream.cjs`:
- Around line 186-193: 将 buildSseFrames(responseBytes, responseMode)
从每次请求的处理流程移到模块加载阶段，仅构建一次并缓存结果；在 writeNext 及其所属请求逻辑中复用该模块级
frames，保持现有发送顺序与计数行为不变。
In `@tests/load/issue-1408-replay-oom/run-wave.sh`:
- Around line 100-118: 更新 TTL 轮询中 inspect-redis.cjs 的错误处理，将 stderr
重定向到独立临时文件而不是丢弃，并在超时时输出该文件内容。为 CCH_TTL_CLEANUP_TIMEOUT_SECONDS 和
CCH_RDB_DELAY_SECONDS 增加整数格式校验，在进入超时计算或 sleep 前拒绝无效值并返回清晰错误。
🪄 Autofix
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 9c674011-3aba-4215-8cec-5b7e81d6a48a
📒 Files selected for processing (29)
.env.exampledocs/research/issue-1415-session-response-body-dedup.mddocs/troubleshooting/issue-1408-replay-oom.mdsrc/app/v1/_lib/proxy/response-handler.tssrc/app/v1/_lib/proxy/warmup-guard.tssrc/lib/config/env.schema.tssrc/lib/session-manager-detail-snapshots.test.tssrc/lib/session-manager-response-body-dedup.test.tssrc/lib/session-manager.tstests/configs/integration.config.mtstests/integration/billing-model-source.test.tstests/integration/session-response-body-dedup-redis.test.tstests/load/issue-1408-replay-oom/README.mdtests/load/issue-1408-replay-oom/drive-disconnect-waves.cjstests/load/issue-1408-replay-oom/inspect-redis.cjstests/load/issue-1408-replay-oom/mock-upstream.cjstests/load/issue-1408-replay-oom/run-wave.shtests/load/issue-1408-replay-oom/sample-container.shtests/load/issue-1408-replay-oom/start-mock-container.shtests/unit/lib/env-store-session-response-body.test.tstests/unit/load/issue-1408-mock-upstream.test.tstests/unit/proxy/pricing-no-price.test.tstests/unit/proxy/response-handler-abort-listener-cleanup.test.tstests/unit/proxy/response-handler-client-abort-drain.test.tstests/unit/proxy/response-handler-endpoint-circuit-isolation.test.tstests/unit/proxy/response-handler-gemini-stream-passthrough-timeouts.test.tstests/unit/proxy/response-handler-lease-decrement.test.tstests/unit/proxy/response-handler-non200.test.tstests/unit/proxy/warmup-guard.test.ts
🧪 测试结果
总体结果: ✅ 所有测试通过 |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review Summary
No significant issues identified in this PR after reviewing the new Redis response-body dedup path, the rollout compatibility logic, and the added Redis/load-test coverage.
The main runtime changes stay aligned with the existing session snapshot consumers, and I did not find a diff-scoped defect that cleared the reporting threshold after validation.
PR Size: XL
- Lines changed: 2363
- Files changed: 29
- Split suggestions: If follow-up work is needed, consider separating runtime dedup logic, Redis integration coverage, and load-fixture/docs evidence updates into smaller PRs.
Review Coverage
- Logic and correctness - Clean
- Security (OWASP Top 10) - Clean
- Error handling - Clean
- Type safety - Clean
- Documentation accuracy - Clean
- Test coverage - Adequate
- Code clarity - Good
Automated review by Codex AI
There was a problem hiding this comment.
- Reviewed PR
#1417across the requested perspectives, focused on the diffed runtime changes, rollout logic, and added Redis/load-test coverage. - Applied the
size/XLlabel and posted the PR review summary viagh. - I did not find a diff-scoped issue that cleared the reporting threshold after full-context validation, so I submitted a no-findings summary and included split suggestions because the PR is XL.
- I also tried to run targeted Vitest coverage for the new session-body paths, but this runner does not have
bunxor a localvitestbinary available.
There was a problem hiding this comment.
Code Review Summary
This PR implements Redis response body deduplication by collapsing legacy/before/after body views into a single request-scoped Redis Hash with atomic Lua-based generation switches. The implementation is architecturally sound: the Lua scripts have correct ARGV mappings, backward compatibility is carefully preserved through legacy key fallback in the reader, and the two-stage rollout (SESSION_RESPONSE_BODY_DEDUP_ENABLED) cleanly coordinates mixed-writer scenarios during rolling deploys.
PR Size: XL
- Lines changed: 2363 (2195 additions, 168 deletions)
- Files changed: 29 (5 source files, 24 test/doc/load-test files)
Issues Found
| Category | Critical | High | Medium | Low |
|---|---|---|---|---|
| Logic/Bugs | 0 | 0 | 0 | 0 |
| Security | 0 | 0 | 0 | 0 |
| Error Handling | 0 | 0 | 0 | 0 |
| Types | 0 | 0 | 0 | 0 |
| Comments/Docs | 0 | 0 | 0 | 0 |
| Tests | 0 | 0 | 0 | 0 |
| Simplification | 0 | 0 | 0 | 0 |
No issues met the reporting threshold (confidence >= 80).
Key verification areas
- Lua ARGV/index mapping: Verified all three scripts (
write:v1,write-legacy:v1,read:v1) against their JS call sites. KEYS and ARGV offsets are correct for both write paths (4 keys, 10+ ARGV) and the read path (2 keys, 1 ARGV). - Backward compatibility: The reader Lua falls back to legacy
GETwhen the bundle key does not exist (EXISTS == 0), so sessions written by pre-bundle instances remain readable. Thelayout=legacybranch also reads from the legacy SETEX key, ensuring consistency whenDEDUP_ENABLED=false. - Over-budget semantics: When aggregate unique bytes exceed the limit, all bodies are dropped and
over_budget=1is written withpresent:*flags but no refs. The reader correctly returnsbody: nullfor this case. The behavior difference ingetSessionResponsePhaseSnapshot's null guard (returns a null-field snapshot instead ofnullwhenexists && present) is harmless —normalizeResponseSnapshotinactive-sessions.tshandles bothnulland{ body: null, ... }identically. - Error handling: All catch blocks log with appropriate context. The
normalizeSessionResponseBodycatch (non-JSON bodies like SSE streams) mirrors the existing pattern instoreSessionResponse. No silent failures. - Atomicity: DEL+HSET within a single Lua eval is atomic. The reader's EXISTS+GET/HGET sequence is also atomic within one eval, preventing marker/body splits during generation switches.
- Test coverage: Unit tests cover all-identical, pair-identical, all-distinct, post-redaction, UTF-8 limits, over-budget markers, legacy fallback, retries, rollout switches, and disabled storage. Integration tests cover real Redis Hash storage, retry generations, rollback/forward rollout, and mixed writers.
Review Coverage
- Logic and correctness - Clean
- Security (OWASP Top 10) - Clean
- Error handling - Clean
- Type safety - Clean
- Documentation accuracy - Clean
- Test coverage - Comprehensive
- Code clarity - Good
Automated review by Claude AI
Add a session-level response body generation counter and a per-request generation snapshot taken when a request sequence is issued. The dedup and legacy Lua writers now check that the request-scoped generation matches the current session generation before writing, preventing a late writer from recreating response bodies after terminateSession has run. Full termination advances the session generation atomically via a new Lua script that also deletes every bundle registered in the session ZSET index, their legacy view keys, and the sequence-1 fallback keys. Provider-scoped termination is unchanged and does not touch shared session artifacts. Requests carrying a keyId but no valid sequence now fail closed instead of falling through to the legacy compatibility path, avoiding any bypass of the termination fence.
Pre-cache SSE frames once at mock-upstream startup instead of rebuilding them per request, and always swallow completion rejections in the driver so complete-mode failures surface through the controlled receipt timeout rather than as unhandled rejections. run-wave.sh now validates that CCH_RDB_DELAY_SECONDS and CCH_TTL_CLEANUP_TIMEOUT_SECONDS are non-negative integers, uses separate temp files for TTL inspection stdout/stderr so diagnostics are preserved on timeout, and cleans up temp files via an EXIT trap with explicit INT and TERM handlers. The unit fixture test now strips all CCH_ environment variables before applying test overrides, adds coverage for cached-frame reuse, complete-mode receipt-timeout failure reporting, invalid integer rejection, and TTL timeout diagnostic preservation.
| pipeline.eval( | ||
| DELETE_SESSION_RESPONSE_BODY_BUNDLES_LUA, | ||
| 5, | ||
| buildSessionResponseBodyBundleIndexKey(sessionId), | ||
| buildSessionResponseBodyGenerationKey(sessionId), | ||
| `session:${sessionId}:response`, | ||
| buildLegacySessionResponseBodyViewKey(sessionId, 1, "before"), | ||
| buildLegacySessionResponseBodyViewKey(sessionId, 1, "after"), | ||
| SessionManager.SESSION_TTL, | ||
| crypto.randomUUID() | ||
| ); |
There was a problem hiding this comment.
When the response-body Lua command fails after the binding has been terminated, terminateSession ignores that pipeline-command error and still returns success because bindingTerminated is true or another deletion succeeded, leaving the multi-megabyte response-body bundles in Redis until their TTL expires.
Knowledge Base Used: Redis Caching and Session Tracking
Prompt To Fix With AI
This is a comment left during a code review.
Path: src/lib/session-manager.ts
Line: 3659-3669
Comment:
**Bundle cleanup failures report success**
When the response-body Lua command fails after the binding has been terminated, `terminateSession` ignores that pipeline-command error and still returns success because `bindingTerminated` is true or another deletion succeeded, leaving the multi-megabyte response-body bundles in Redis until their TTL expires.
**Knowledge Base Used:** [Redis Caching and Session Tracking](https://app.greptile.com/ygxz/-/custom-context/knowledge-base/ding113/claude-code-hub/-/docs/redis-caching-and-sessions.md)
---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ef603e8048
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| local generation = redis.call('GET', KEYS[2]) | ||
| if not generation then | ||
| generation = '0' | ||
| redis.call('SETEX', KEYS[2], ARGV[2], generation) |
There was a problem hiding this comment.
When a streaming request remains open longer than SESSION_TTL (300 seconds by default), the global generation set here and the request generation set at line 696 both expire before the response finalizer writes its body. WRITE_SESSION_RESPONSE_BODY_BUNDLE_LUA requires both keys to exist and match, so it silently returns 0 and the completed response body is never persisted; this affects streams the response handler explicitly keeps alive beyond SESSION_TTL (response-handler.ts:406-412). Refresh these generation keys with the streaming heartbeat or otherwise retain them for the full request lifetime while preserving the termination fence.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/research/issue-1415-session-response-body-dedup.md`:
- Around line 115-121: 明确 requestOwnerKey 的一致性边界：若其 TTL 属于 dedup 原子写入契约，更新
storeSessionResponseBodySet 及对应 Lua 脚本，将 requestOwnerKey 的 TTL
刷新纳入同一原子执行；否则收窄并说明该契约不包含 requestOwnerKey，仅覆盖 Hash、ZSET 和 legacy key。
🪄 Autofix
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: e7d2776f-8b3e-4557-9b2a-432cc4b2af88
📒 Files selected for processing (11)
.env.exampledocs/research/issue-1415-session-response-body-dedup.mdsrc/lib/session-manager-detail-snapshots.test.tssrc/lib/session-manager-response-body-dedup.test.tssrc/lib/session-manager.tstests/integration/session-response-body-dedup-redis.test.tstests/load/issue-1408-replay-oom/drive-disconnect-waves.cjstests/load/issue-1408-replay-oom/mock-upstream.cjstests/load/issue-1408-replay-oom/run-wave.shtests/unit/lib/session-manager-terminate-session.test.tstests/unit/proxy/issue-1408-load-fixture.test.ts
🚧 Files skipped from review as they are similar to previous changes (4)
- src/lib/session-manager-detail-snapshots.test.ts
- .env.example
- tests/load/issue-1408-replay-oom/run-wave.sh
- tests/load/issue-1408-replay-oom/mock-upstream.cjs
| dedup 写入 Lua 同时声明 Hash, 三个旧正文 key 和 session-level bundle index: | ||
|
|
||
| 1. 校验 request owner 和 request/session response body generation 一致. | ||
| 2. 原子删除上一 generation 的 Hash 和旧正文 key. | ||
| 3. 写入固定字段, present/ref 和唯一 body. | ||
| 4. 对整个 Hash 和 generation marker 执行 `EXPIRE`. | ||
| 5. 按 Redis `TIME` 的毫秒过期分数剪枝 ZSET, 登记当前 bundle, 并刷新 index TTL. |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
rg -n -i -C 12 \
'storeSessionResponseBodySet|evalsha|eval|zadd|zremrangebyscore|expire|hset|del|unlink' \
src/lib/session-manager.ts testsRepository: ding113/claude-code-hub
Length of output: 50380
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- candidate files ---'
git ls-files | rg '(^|/)(session-manager|redis|session).*\.(ts|tsx|js|lua)$|issue-1415-session-response-body-dedup\.md$' | head -200
printf '%s\n' '--- exact symbols ---'
rg -n -S -C 8 \
'storeSessionResponseBodySet|session-response|response.?body|legacy|zremrangebyscore|EVALSHA|evalsha|\.eval\(|\.multi\(|HSET|ZADD|EXPIRE|expire' \
src tests \
-g '*.ts' -g '*.tsx' -g '*.js' -g '*.lua' \
| head -1000Repository: ding113/claude-code-hub
Length of output: 50379
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- tracked candidate files ---'
git ls-files | rg -i 'session|redis' | rg '\.(ts|tsx|js|lua)$' | head -300
printf '%s\n' '--- exact symbol locations ---'
rg -n -S 'storeSessionResponseBodySet|storeSessionResponseBody|responseBodyGeneration|legacy.*key|bundle.*index' src tests \
-g '*.ts' -g '*.tsx' -g '*.js' -g '*.lua' || true
printf '%s\n' '--- relevant Redis calls in session manager ---'
rg -n -S -C 18 'evalsha|eval|zadd|zremrangebyscore|expire|hset|del|unlink|multi|pipeline' \
src/lib/session-manager.ts 2>/dev/null || trueRepository: ding113/claude-code-hub
Length of output: 50379
🏁 Script executed:
#!/bin/bash
set -euo pipefail
files="$(git ls-files | rg -i 'session|redis' | rg '\.(ts|tsx|js|lua)$' || true)"
printf '%s\n' "$files" | head -200
printf '%s\n' '--- matches across source ---'
rg -n -S -C 6 \
'storeSessionResponseBodySet|responseBodyGeneration|zremrangebyscore|legacy.*key|bundle.*index|evalsha|eval\(' \
src -g '*.ts' -g '*.tsx' -g '*.js' -g '*.lua' || trueRepository: ding113/claude-code-hub
Length of output: 50380
🏁 Script executed:
#!/bin/bash
set -euo pipefail
python3 - <<'PY'
from pathlib import Path
import re
patterns = re.compile(
r"storeSessionResponseBodySet|responseBodyGeneration|"
r"zremrangebyscore|zadd|hset|expire|unlink|evalsha|"
r"redis\.eval|\.eval\("
, re.I
)
for p in Path(".").rglob("*"):
if not p.is_file() or any(part in {"node_modules", ".git", "dist", "build"} for part in p.parts):
continue
if p.suffix not in {".ts", ".tsx", ".js", ".lua"}:
continue
try:
lines = p.read_text(errors="replace").splitlines()
except OSError:
continue
hits = [i for i, line in enumerate(lines) if patterns.search(line)]
if not hits:
continue
print(f"FILE {p} HITS {len(hits)}")
for i in hits[:80]:
lo, hi = max(0, i-2), min(len(lines), i+3)
for j in range(lo, hi):
print(f"{p}:{j+1}:{lines[j]}")
print("--")
PYRepository: ding113/claude-code-hub
Length of output: 50380
🏁 Script executed:
#!/bin/bash
set -euo pipefail
repo='ding113/claude-code-hub'
for path in \
src/lib/session-manager.ts \
tests/unit/lib/session-manager.test.ts \
tests/unit/session-manager.test.ts
do
url="https://raw.githubusercontent.com/$repo/main/$path"
body="$(curl -LfsS "$url" 2>/dev/null || true)"
if [ -n "$body" ]; then
printf '%s\n' "--- $path ---"
printf '%s\n' "$body" | rg -n -S -C 10 \
'storeSessionResponseBodySet|responseBodyGeneration|zremrangebyscore|zadd|hset|expire|unlink|evalsha|eval\(' \
| head -800 || true
fi
doneRepository: ding113/claude-code-hub
Length of output: 5821
🌐 Web query:
site:github.com/ding113/claude-code-hub "storeSessionResponseBodySet" OR "responseBodyGeneration"
💡 Result:
The terms storeSessionResponseBodySet and responseBodyGeneration are not standard API methods or formal technical concepts in general programming or the referenced "claude-code-hub" codebase [1][2][3][4]. Research indicates that the codebase associated with these terms uses explicit session management functions such as storeSessionResponse, storeSessionUpstreamResponseMeta, and storeSessionRequestBody [1][2][3]. Based on internal documentation and issue discussions within the claude-code-hub repository: 1. Session Data Handling: Request and response bodies are currently managed primarily as temporary data in Redis with a Time-To-Live (TTL) (typically 5 minutes), rather than being persisted in the primary database [3][5]. 2. Persistence Logic: Feature requests and architectural discussions confirm that there is no native storeSessionResponseBodySet method [5][6]. Developers have explored extending the database schema (e.g., adding request_body or response_body columns to a message_request table) to support persistence, but this is a custom implementation requirement rather than an existing API [5]. 3. Response Processing: Code related to response handling often focuses on streaming (SSE), filtering, and metadata extraction (such as actual-response-model.ts) rather than setting a generic "body generation" state [7][6][4]. If you encountered these specific terms in a codebase, they are likely internal custom identifiers, deprecated function names, or possibly typos for functions like storeSessionResponse [2][3]. It is recommended to search the local codebase directly for these strings to identify their specific implementation and context.
Citations:
- 1: https://github.com/ding113/claude-code-hub/blob/2b60af9f/tests/unit/proxy/session-guard-warmup-intercept.test.ts
- 2: https://github.com/ding113/claude-code-hub/blob/2b60af9f/tests/unit/proxy/response-handler-non200.test.ts
- 3: e6a3b5c
- 4: f4f161c
- 5: 希望将请求体落盘 #971
- 6: 探讨：新增响应过滤器的功能 #783
- 7: fix(proxy): finalize completed responses streams after client abort #1251
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- writer scripts and invocation ---'
sed -n '40,180p' src/lib/session-manager.ts
sed -n '2380,2525p' src/lib/session-manager.ts
printf '%s\n' '--- response-body key mutations outside writer method ---'
rg -n -S -C 5 \
'response-bodies:v1|response-body-bundles:v1|response-body-generation:v1|snapshot:response:(before|after):body|:response`|:response"' \
src/lib/session-manager.ts src/lib/redis src/app/v1/_lib/proxy tests/integration/session-response-body-dedup-redis.test.ts \
-g '*.ts' | head -1200Repository: ding113/claude-code-hub
Length of output: 20996
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- body-set remainder ---'
sed -n '2470,2605p' src/lib/session-manager.ts
printf '%s\n' '--- owner refresh implementation and callers ---'
rg -n -S -C 18 \
'refreshSessionRequestOwner|REQUEST_OWNER|:owner|request owner' \
src/lib/session-manager.ts src/lib/redis tests \
-g '*.ts' | head -1200Repository: ding113/claude-code-hub
Length of output: 50379
将 request owner 的 TTL 刷新纳入原子写入，或收窄原子性契约。
两个写入 Lua 脚本已在同一次执行中完成 HSET、legacy key 清理、ZREMRANGEBYSCORE、ZADD 和 bundle/index EXPIRE。storeSessionResponseBodySet 仍先通过独立的 SETEX 刷新 requestOwnerKey。如果该 TTL 属于原子写入契约，请将刷新移入 Lua；否则明确该 key 不属于 Hash、ZSET 和 legacy key 的一致性范围。
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/research/issue-1415-session-response-body-dedup.md` around lines 115 -
121, 明确 requestOwnerKey 的一致性边界：若其 TTL 属于 dedup 原子写入契约，更新
storeSessionResponseBodySet 及对应 Lua 脚本，将 requestOwnerKey 的 TTL
刷新纳入同一原子执行；否则收窄并说明该契约不包含 requestOwnerKey，仅覆盖 Hash、ZSET 和 legacy key。
github-project-automation
Bot
moved this from Backlog
to Done
in Claude Code Hub Roadmap
2 participants
Summary
STORE_SESSION_RESPONSE_BODY=falsebehavior.SESSION_RESPONSE_BODY_DEDUP_ENABLED=falsefor a reader-first two-stage rollout. In dedup mode,SESSION_RESPONSE_BODY_MAX_BYTESbounds aggregate unique UTF-8 body bytes per request.Problem
A single request's response body was written to up to three independent Redis keys (legacy response, before snapshot body, after snapshot body). When the three views held identical or overlapping content, Redis stored up to three full copies. #1414 capped each copy at 5 MiB, but the 3x amplification remained - the same amplifier behind the #1408 OOM incidents, where 64 completed SSE responses produced roughly 1.4 GiB of raw body values and pushed Redis peak to ~1.5 GiB during the RDB save window. This PR collapses the three views into one request-scoped Redis Hash so identical post-redaction bodies share a single physical value.
Related Issues
SESSION_RESPONSE_BODY_MAX_BYTES; this PR repurposes it from a per-copy cap to an aggregate unique-body-byte budget in dedup modeSTORE_SESSION_RESPONSE_BODY=trueTests and verification
64 x 5 MiB acceptance
body:*fields totaling 335,544,320 bytes, exactly64 x 5,242,880.used_memory_peakwas 437,018,776 bytes; Redis cgroup peak was 376,094,720 bytes; app cgroup peak was 417,484,800 bytes.BGSAVEcompleted withrdb_last_bgsave_status=ok; Redis remained running withOOMKilled=falseandExitCode=0.SESSION_TTL, the same manifest had zero bundles and zero legacy response body keys.Acceptance: https://app.lobehub.com/acceptance/d29406fe-a395-45bd-9b72-2229fa4d40c1
Final round: https://app.lobehub.com/verify/1f245416-a59f-40f2-bf45-9f345b18cea4
Description enhanced by Claude AI
Greptile Summary
The PR consolidates each request’s legacy, before, and after response-body views into a Redis hash with generation-fenced reads, writes, rollout compatibility, and termination cleanup.
Confidence Score: 4/5
The PR is not yet safe to merge because full session termination can report success while the newly indexed response-body bundles remain undeleted.
The response-body cleanup is executed as one pipeline command, but its command-level error is ignored and a successful binding termination or unrelated deletion still makes terminateSession return true, leaving the memory-heavy artifacts until TTL expiry.
Files Needing Attention: src/lib/session-manager.ts
Important Files Changed
Sequence Diagram
Prompt To Fix All With AI
Reviews (2): Last reviewed commit: "test(load): stabilize replay-OOM fixture..." | Re-trigger Greptile
Context used: