e0f22f808fe8b3f9159a8cc76c41915fef4dd524
4
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
  samiandClaude Sonnet 5
|
322a20efa5 |
core: fix SegmentBudget over-admission and add a wait-list wakeup
Root cause of the M7 RSS gap (core/docs/m7-baseline.md: ~70 MB measured against a 60 MB target): SegmentBudget::confirm_slot() only checked a task's own held count against its own target -- never the engine-wide active_ sum. reallocate_locked()'s two-pass fairness allocation does bound sum(target) <= max_active_ at the moment it computes a plan, but that bound says nothing about sum(held): a task can be legitimately holding more than its own just-lowered target for a while (yield is deferred to a segment boundary, never mid-segment -- ADR 0011 A1), and another task's target can correctly rise to claim that capacity before the first task has physically released it. Both confirm_slot() calls could then succeed against their own, individually-correct targets while sum(held) exceeded max_active_ -- tools/bench heap-profile caught this directly: budget.active reading 56-86 against a total of 32. confirm_slot() now also checks active_ < max_active_, unconditionally, as a backstop that doesn't depend on any task's target bookkeeping being in sync with what every other task holds. That creates a liveness question the original design never answered: a task denied only by this new check has a target that's already correct, so it never changes again and reallocate_locked()'s plain "fire a callback when a task's target changes" mechanism never revisits it. Task gained a waiting_for_slot flag, set on exactly this denial; release_slot()/deregister_task() (the only two places that free real capacity) now hand a freed slot directly to the highest-priority waiting task via wake_one_waiter_locked(), if reallocate_locked()'s own plan didn't already produce a callback for anyone. download_task.cpp's fill_slots_locked() needed a matching fix: a woken task's stalled segments (SegState::stalled -- backed off mid-retry, its own release_slot() already called) have no live worker and never surface through Segmenter::assign_slot(), which only hands out unassigned or fresh ranges. fill_slots_locked() now restarts any stalled segment with no live worker directly (bounded by slot_target, same as its assign_slot() loop) before looking for new work; a segment it doesn't get to keeps its own scheduled retry_worker() timer as a second chance. Also fixes a real TSan-caught data race this work surfaced: SegWorker:: speed_bps was written only by its own segment's curl callback and, before Progress.speed_bps's polled-path fix, only ever read from that same thread -- safe without synchronization. snapshot_progress() reading it from whatever thread calls DownloadHandle::progress() broke that invariant (workers_mu's shared_lock protects the workers map's structure, not an individual SegWorker's fields). Now std::atomic<double> with relaxed ordering -- an informational EMA, nothing synchronizes real state on it -- rather than adding a lock to the write side. core/tests/segment/budget_test.cpp adds two tests reproducing the actual gap (budget_active_never_exceeds_max_active_segments_under_concurrent_load, budget_wait_list_wakes_a_task_whose_target_never_changed) plus a sanity baseline (budget_release_wakes_a_denied_waiter), and introduces AsyncFakeTask + TestTimer for the one existing test that drives the budget from multiple concurrent threads -- mirroring production's real dispatch (register_task()'s on_target lambda posts through host.schedule(), download_task.cpp, never a synchronous call) rather than adding reentrancy-guarding machinery to SegmentBudget itself to compensate for a synchronous test double being unlike production. See docs/adr/0017 for the full writeup, including what an earlier version of this fix got wrong chasing a same-thread reentrancy hazard that doesn't actually exist in production. core/docs/m7-baseline.md updated: the RSS number now clears the DoD line (45.41 MiB via heap-profile), root-caused rather than just re-measured. Co-Authored-By: Claude Sonnet 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Q3QrF7rCt21bkAjt9BCDFQ |
||
|
samiandClaude Sonnet 5
|
91636f8a4d |
core: add the download engine — task machine, DownloadHandle, Engine
Stage 8 of the CORE build order: the bodies behind the DownloadSpec / callback API reviewed in core/docs/engine-api-m1.md. Wires probe -> segment workers -> WriteBuffer -> SparseFile -> .veloxpart.meta -> retry/backoff -> SegmentBudget -> RateLimiter -> callbacks into one event-driven machine. - Engine (src/engine.cpp): owns HttpClient, Prober, SegmentBudget, RateLimiter and one timer jthread (min-heap of scheduled fns). start() builds a task and returns a DownloadHandle; ~Impl quiesces every task before joining the timer so no callback fires during teardown. - DownloadTaskState (src/task/download_task.cpp): one `mu` task lock; a shared_mutex over the worker map for the curl write path; callbacks collected under `mu` and fired after release via a separate deferred queue; weak_from_this() in every async hop. State machine over the CORE-owned EngineState subset, auto-pause on 401/407 and on a 200 where 206 was expected, validated resume via If-Range. - digest (src/task/digest.cpp): OpenSSL EVP hash_file() for the optional post-download checksum; links OpenSSL::Crypto PRIVATE. - Segmenter::release_segment(): hand a paused segment back to the pool unassigned so resume's assign_slot() picks it up instead of splitting a still-"assigned" range and orphaning its front half. - DownloadHandle now names the real control block (vdm::task:: DownloadTaskState, defined only in the engine TU) via a namespace-scope fwd decl and a public-but-effectively-engine-only ctor, replacing the nested State/friend pair. Every public signature is unchanged; DAEMON (vdm-79) confirmed sched/ names only the public API. Fixes found while building the end-to-end suite (tests/task/engine_test.cpp, 9 cases against tools/testserver, green under ASan/UBSan and TSan): - a dropped connection lost its unflushed WriteBuffer tail while advance() had already counted those bytes as done -> a retry resumed past an unwritten hole. Flush on the failure path. - when the byte counters hit total while other workers were still live, teardown dropped their buffered tails. Now: cancel them and let each worker's own seg_finished drain it (the `assembling` state), last one starts verification -- no cross-thread buffer access. - seg_head() let a 401 with credentials present abort before libcurl's resend; now it proceeds once and acts on the final status. Co-Authored-By: Claude Sonnet 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01HPPSGhiArbvQgwC2DNiURS |
||
|
samiandClaude Sonnet 5
|
5d81b4cdae |
core: segment/segmenter + segment/budget (stage 6)
vdm/ids.hpp — TaskId, an opaque engine handle (DAEMON keeps the wire UUID <-> TaskId map; the engine never sees the UUID). segment/segmenter — per-download range management (docs/04 §3). Initial lazy split; assign_slot() splits the largest remaining range when the budget grants a slot; on_complete(may_steal) either *steals* the second half of the largest remaining range for the same worker (slot-neutral) or returns nullopt so the caller *yields* the slot (ADR 0011 A1); on_failed() returns requeue only on the 3rd consecutive connection error with a mirror present — the remaining range is orphaned and re-split. Non-resumable or unknown-size => exactly 1 segment; never split below min_segment_bytes (1 MiB). Resume ctor rebuilds from a persisted table (falls back to a fresh layout if it doesn't tile [0,total)). One mutex == "the task lock"; segment fields are std::atomic and the store is a std::deque so a steal's append never moves a worker's record. segment/budget — the global allocator (ADR 0011). Owns exactly one ceiling (maxActiveSegments) and min-1-before-seconds fairness: a two-pass allocation (guarantee pass gives every wanting task 1 slot in DAEMON's priority order, then a growth pass round-robins the rest up to each task's effective cap = min(per_task_cap, host cap, 1 if non-resumable)), recomputed from scratch on every edge so a live set_max_active_segments cut naturally yields the excess lowest-priority-first, never a mid-segment kill. DAEMON-facing surface exactly as promised in daemon/docs/core-requests-m1.md / ADR 0011: budget(), segments_active(), starved_tasks(), starved_since(), set_max_active_segments (drain), set_host_segment_cap, set_task_order, on_budget_changed (a jthread coalesces at <=4 Hz; the tasks_starved 0<->nonzero edge fires immediately). Callbacks are copied out and run after the lock is dropped. Tests: segmenter split/steal/requeue/resume math + a concurrent steal-and-advance run; budget min-1 under a tight budget, round-robin growth, host-cap and non-resumable clamps, live-lower shedding lowest-priority-first, starvation below the task count, starved-edge notification, and a concurrent set_want hammer. Green under ASan/UBSan; the steal path and the budget green under TSan. Co-Authored-By: Claude Sonnet 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01HPPSGhiArbvQgwC2DNiURS |
||
|
samiandClaude Opus 5
|
8bb683b09d |
scaffold: project structure, wire contract, roadmap and agent briefs
Lays out Velox Download Manager (IDM-class download manager for Ubuntu 26.04) as a monorepo ready for parallel lane development. No implementation code by design. - docs/: architecture, roadmap M0-M7, IDM-parity GUI spec, engine design, Firefox extension spec, risks/spikes, packaging - contracts/: wire-contract skeleton (JSON Schema + fixture templates) — the single synchronization point between lanes - docs/agents/: one brief per lane (PROTO, CORE, DAEMON, GUI, EXT, PKG/QA) with owned directories, build order and definition of done - CLAUDE.md: rules of engagement — lane ownership, layering, non-negotiables - CMake scaffolding with dev/tsan/release/ci presets Two environment findings shape the design: Firefox here is the Mozilla snap (native-messaging risk, so the extension carries a loopback-WebSocket fallback), and Wayland forbids passive clipboard monitoring (so clipboard capture is explicit-action-first). Co-Authored-By: Claude Opus 5 <[email protected]> |