@bighead_datou

⚙️ 交易撮合技术 | ⚡ HFT | ⏱️ Ultra Low Latency | 🏋️ 重训 | 👷 劳工 in Singapore 🇸🇬 | I have a big head 🤓

 Singapore
Joined April 2012
  • Tweets457
  • Following393
  • Followers195
42Media
All right! 我，Big Head，技术能力平平、对很多事情一知半解，却总能莫名其妙地升职，发财，成功，朋友牛逼，好玩的事有趣的人络绎不绝。 Yeah. “Fucking billionaires!”
7
2
27,797
真发钱啊，但我也用不了X money啊
12
10/21 PHP 发了安全更新，修了两个 critical RCE（CVE-2026-109618 / 109619）。官方 advisory 对第二个的描述大意是：worker 里的全局状态还指向上一个 request 已释放的内存，正常 reset 被 fatal error 跳过了，下一个 request 在正常初始化之前就用上了它，形成 use-after-free。 这类问题不只 PHP 有。任何「长生命周期 worker + 对象复用」的系统都可能踩到，连接池、游戏服务器、各种为了躲 GC 做的对象池都算。 对象复用的坑，往往不在 happy path： - reset 写在 release 里，某个异常分支跳过了 reset 却照样把对象还回池子 - 下一次 acquire 拿到它，大部分字段被覆盖了，偏偏一个可选字段没被覆盖，带着上一次请求的值往下走 - 测试全绿，因为测试数据里每个字段都填了 第一个 CVE 据 advisory 描述也是同一类：内存还没初始化完就被 OOM 打断，request shutdown 又去读这块半初始化的数据。 比较稳的做法：reset 放在 acquire 侧，默认池子里的对象是脏的；错误路径和正常路径走同一个 cleanup；再留一个 debug 模式，release 时把对象填满垃圾值，让漏 reset 的字段在测试里直接炸出来。 我认为就是一种防御性编程吧
9
交易系统最怕的，不是慢，而是你根本不知道它什么时候会突然变慢。 平时延迟稳稳地跑在几十微秒，结果一看 P99.9，直接飙到几毫秒，甚至更高。 最头疼的是，这种长尾延迟往往不是某一行代码慢，而是各种因素叠加出来的。 CPU cache miss、TLB miss、线程抢占、NUMA 远程访问，甚至偶发的 page fault。 单独看可能都不严重，但几个因素撞到一起，尾延迟就很难看了。 更烦的是，压测数据挺漂亮，一上生产，流量一突增，P99.9 就开始乱跳，想复现都难。 所以现在做性能优化，我更关心 latency distribution，而不只是 average latency。 低延迟系统真正难的，不是再快几微秒，而是搞清楚最慢的那 0.1% 到底发生了什么。
16
LGTM
Rod Pers
@srepdorOct 19
Took this one step further. Vibe coded my own TV OS with Claude Code No more ads: replaced with a live 3D model of my apartment. Lights + ceiling fan wired in, they turn on in the 3D in real time. The TV inside the 3D shows the live TV 🤯 What should I add next?
1
25
A matching engine can look fast if the latency clock starts after dequeue. During a burst, orders may spend more time waiting than matching. Measure from ingress, and keep queueing time visible. The latency you leave out still gets paid by the user.
18
In a replicated matching engine, what does an order ACK promise? Received, sequenced, and committed to a quorum are different milestones. If the leader dies right after the ACK, can the order disappear? That answer should shape your retry and deduplication design.
20
1/ 经典重读：LMAX Disruptor（2011）。很多人记住的是「环形队列」，但这套设计快，靠的不是环。
1
43
more replies
8/ Fowler 文中转述了 LMAX 团队反复强调的一点，到今天仍然值得记住："coming up with meaningful performance tests is often harder than developing the production code"。
1
8
Disruptor 论文：lmax-exchange.github.io/disr… Martin Fowler《The LMAX Architecture》：martinfowler.com/articles/lm…
9
交易所你用过几家，它们的撮合引擎怎样工作，你知道多少？
32
微服务也是类似的，似乎聊架构首先就是微服务拆分，而在做交易系统这些年，花了很多时间才把微服务架构改成现在的架构
This quoted post is unavailable.
Try viewing anyway · View on X
1
33
Big Head: "Three more classes, and I get my boat license." Richard: "How many classes does it take?" Big Head: "Three." The energy I bring to every new year's resolution. Progress: 0%. Confidence: 100%. 💪 （sent by Grok bot）
10
Big Head, S2E9: "Gavin promoted me just to spite you. And now he keeps promoting me for some inexplicable reason." Career goal unlocked: be the inexplicable reason. 🥤 （sent by Grok bot）
10
Hooli brogrammer: "Do you have any skill at all other than magically failing your way to the top?" Big Head: "I have a boat. I guess that's not really a skill." Me: no boat. Just a transit card and unearned confidence. 🚇 （sent by Grok bot）
10
Load more