How do I feel about breaking through 30 million?

Watching one by one "celebrating 30 million" operation promotion activities, but the heart is not little excitement, but become the starting point of loss.

A "30 million" behind a lot of icy numbers: 100+ (average 84GB memory +16c) Server (physical machine + virtual machine), 50+ application services, 300+ Redis instance, the application process, 10w+ Peak concurrency, 800 million + Daily API visits, daily 1tb+ system logs ..., looks like I'm faking the rhythm.

What you see is numbers, what you can't see is the truth.

If the system is a partial query class application, I think a few claims that 10W concurrent Redis instance enough to support tens users +10w concurrency-scale system, you will feel NB? If 50% of the system logs in 1TB are non-active, and maybe only 0.1% of the logs have analytical value, do you still think NB? If 30 million users have 99.99% of the users are "forced" to use, you will feel NB ...

Throughout the years, the most exciting thing is that when a few colleagues in front of the server 7*7*24 hours of uninterrupted search for problems and restart service server, often the most time is spent in a haystack, and this "needle" is often a line of code or a configuration of things, but also the "needle" The discovery has put my programmer's career excitement index on the peak. This kind of programmer "simple", I think only programmers can understand. As a programmer, "high performance, high availability, high flexibility" is my goal, as a manager, "political, social, economic" is my top priority (consider the weight of factors in order of priority). I like and am good at simplifying a complex thing, but sometimes I have to face the reality of having to complicate something simple. Programmers are facing forward challenges, and more of the time is a reluctant negative mood.

Today's figures are actually just a four-year extension of the results.

According to the current resource size, not to mention 3000W, and a lot of user capacity can support, this confidence stems from 4 years ago on the structure of the decision-making plan. I do not know where the courage, patted his head shouted "independent research and development." It could be a week of continuous troubleshooting. The reason for the short-circuit of the brain, buried the "control" of the desire. Nearly thousands of lines of core code of the autonomous framework to achieve high constraints and norms of the IOC and AOP, discard the various external middleware high dependence (such as session management, data distribution scheme, etc.), so that the System Building initiative and controllability fully in their own hands. From 8 servers to the current 100 servers, from 5 services to 50 services, or from a single local area network to a hybrid private cloud, all of this has been attributed to the persistence of system controllability and sustainability, recalling these "forward-looking" decisions, I do sometimes feel the pride of the Ness.

I can't guarantee 99.99% stability, but ...

The system has a bug is normal, operation and maintenance error is often the case, the system on-line error rollback, so I do not dare to guarantee the normal operation of the system 100%. Through the "Big Data" analysis of the failure, our system failure rate (all size failure) about 30%, that is, three days a "Nikkei", so I can not guarantee 4 9 stability. However, the data show that 99.99% of the fault is not the internal failure of our system, but the docking of external system failure. No matter how powerful our system is, "the system is busy, please try again later" is our only most valuable downgrade solution. Some friends often give me feedback, why do you use often "system busy", I would be very proud to tell them: "Fortunately we do, for others, give you a white screen, even ' system busy ' do not say to you."

Sometimes a "satisfying" reward is much warmer than these numbers.

At the same time in charge of the large number of people and several projects, in any case, the smooth operation of the whole organization is due to the above years of thinking and tools accumulation, but most of my comfort is not the above system user volume breakthrough, but other small-scale "self-application" to the user experience to bring satisfactory feedback. A string of big data behind may be a cold KPI indicators, but a user's recognition is real gold and silver service feedback. Perhaps, these feelings are only because of the different stages of life in the pursuit of satisfaction have the same idea.