上周去华山医院，需要验血。付费之后，要先在自助机器或者人工窗口刷医保卡登记，系统分配采血窗口，叫号之后才能去抽血。流程设计得挺合理，正常情况下应该很顺畅。
但那天系统坏了。
自助机器插卡没有反应，屏幕转圈不停。走到人工窗口，工作人员面露难色，说系统现在登不进去，请稍等。等候区开始积压人，越来越多，有人站着，有人坐在椅子上刷手机，有人反复走到窗口问"好了吗"。
没有好。
大约过了十几分钟，技术人员来了。两个人，一个坐到工作站前敲命令，一个站在旁边看，身后是一圈患者围观，有人探头看屏幕，有人小声议论。技术人员的后背应该是湿的。
系统的脆弱性
医院的验血登记系统，在整个就诊流程里是一个关键节点：付了费，不登记，就没法抽血；不抽血，后续的诊断就卡住了。这个节点一旦失效，整个流程就断了，没有任何绕过的办法。
这种设计在效率上是合理的——集中管理、统一调度、减少人工失误。但它的代价是：单点故障的影响被放大了。一个系统挂掉，影响的不是某一个窗口，是整个采血区的所有患者。
如果是可以降级的设计——系统挂掉时切换到手工登记模式，哪怕效率低一些——至少流程不会完全停摆。但医院的信息系统历史包袱重，这种弹性往往是奢望。
技术人员的处境
我看着那两个技术人员，有点感同身受。
在公司做IT这么多年，遇到过几次类似的场面：系统宕机，业务停摆，领导和用户都在旁边盯着，而你需要在巨大的压力下冷静排查。这种时候最怕的不是问题本身，而是旁边的人群——他们不是恶意的，只是焦急，但那种无声的压力会让思路变得混乱。
排障需要专注，需要安静，需要一步一步缩小问题范围。但现实给你的是：身后站着二十个人，每隔三分钟有人来问"好了没有"。
他们最后修好了，半小时不到。考虑到那种环境，我觉得其实挺快的。
等待的人群
还有一件事让我印象深刻：整个过程里，患者的情绪虽然焦急，但基本保持克制。有人叹气，有人抱怨，但没有人冲着工作人员发火，也没有人试图强行插队。
大家都知道，这不是某个人的错，是系统出了问题。这种集体的理性，在医院这种本来就容易情绪紧张的地方，还是让我觉得意外地好。
系统修好的那一刻，等候区发出了一阵轻微的骚动，不是欢呼，就是大家同时站起来、准备重新排队的声音。
半小时，就这么过去了。
I went to Huashan Hospital last week for a blood test. After paying, the process requires registering at a self-service machine or staffed counter by tapping your medical insurance card — the system then assigns you a collection window and a queue number. A reasonable design. Under normal circumstances, it flows smoothly.
That day, the system was down.
The self-service machines accepted cards but showed nothing, screens spinning. At the staffed counter, the staff looked apologetic — the system wasn’t letting them in, please wait. The waiting area started filling up. People standing, people scrolling phones in chairs, people walking back to the counter every few minutes to ask if it was fixed yet.
It wasn’t.
About ten minutes in, two technicians arrived. One sat down at a workstation and started typing, the other stood watching. A ring of patients had formed behind them, some craning to see the screen, others murmuring. The technician’s back was probably soaked.
The fragility of the system
The blood test registration system is a critical node in the hospital’s flow: without registering after payment, you can’t be assigned a collection window; without that, you can’t get blood drawn; without that, the downstream diagnosis stalls. Once this node fails, the entire process stops — with no way around it.
The design makes sense from an efficiency standpoint: centralized management, unified scheduling, fewer manual errors. But the cost is that a single point of failure has outsized impact. One system going down doesn’t affect one window — it affects every patient in the entire blood collection area.
A degraded mode — where staff can fall back to manual registration when the system fails, even at lower efficiency — would at least keep the process moving. But hospital information systems carry heavy legacy constraints, and that kind of resilience is often not an option.
The technicians’ situation
Watching those two, I felt a certain recognition.
After years in IT, I’ve been in similar scenes: system down, operations halted, management and users watching over your shoulder, and you need to diagnose the problem clearly under significant pressure. The hardest part in those moments isn’t the technical problem itself — it’s the people around you. They’re not hostile, just anxious. But that silent pressure makes it harder to think.
Troubleshooting needs focus. It needs quiet. It needs methodically narrowing down the possibilities. What you actually get is twenty people standing behind you and someone asking “is it fixed yet” every three minutes.
They got it working in under half an hour. Given the environment, I thought that was pretty good.
The crowd waiting
One more thing stayed with me: throughout the whole episode, the patients were anxious but mostly contained. People sighed. People complained to each other. Nobody directed anger at the staff, and nobody tried to push ahead.
Everyone understood this wasn’t any individual’s fault — it was the system. That collective reasonableness, in a hospital where emotions already run high, was something I didn’t take for granted.
When the system came back up, a low wave of movement passed through the waiting area — not a cheer, just the sound of everyone standing at once, ready to queue again.
Half an hour. That’s all it was.