Well this is suboptimal:
The Java NIO APIs use ByteBuffers as the source and destination of I/O calls, and come in two flavours. Heap ByteBuffers wrap a byte array, allocated in the garbage collected Java heap. Direct ByteBuffers wrap memory allocated outside the Java heap using malloc. Only “native” memory can be passed to operating system calls, so it won’t be moved by the garbage collector. This means that when you use a heap ByteBuffer for I/O, it is copied into a temporary direct ByteBuffer. The JDK caches one temporary buffer per thread, without any memory limits. As a result, if you call I/O methods with large heap ByteBuffers from multiple threads, your process can use a huge amount of additional native memory, which looks like a native memory leak. This can cause your process to unexpectedly run into memory limits and get killed.
Good talk (with transcript) from Paul Biggar about what happened when CircleCI had a massive security incident, and how Jesse Robbins helped them do incident response correctly. ‘On the left, Jesse pointed out that we needed an incident commander. That’s me, Paul. And this is very good, because I was a big proponent, I think lots of were around the 2013 mark, of flat organizational structures, and so I hadn’t really got a handle of this whole being in charge thing. The fact that someone else came in and said, “No, no, no, you are in charge”: extremely useful. And he also laid out the order of our priorities. Number one priority; safety of customers. Number two priority: communicate with customers. Number three priority: recovery of service. I think a reasonable person could have put those in a different order, especially under the pressure and time constraints of the potential company-ending situation. So I was very happy to have those in order. If this is ever going to happen to you, I’d memorize them, maybe put it on an index card in your pocket, in case this ever happens. The last thing he said is to make sure that we log everything, that we go slow, and that we code review and communicate. His point there is that if we’re going to bring our site back up, if we’re going to do all the things that we need to do in order to save our business and do the right thing for our customers and all that, we can’t be making quick, bad decisions. You can’t just upload whatever code is on your computer now, because I have to do this now, I have to fix it. So we set up a Slack channel … This was pre-Slack; it was a HipChat channel, where all of our communications went. Every single communication that we had about this went in that chatroom. Which came in extremely useful the next day, when I had to write a blog post that detailed exactly what had happened and all the steps that we did to fix it and remediate this, and I had an exact time stamps of all the things that had happened.’
Here, we show that deep learning can extract new knowledge from retinal fundus images. Using deep-learning models trained on data from 284,335 patients and validated on two independent datasets of 12,026 and 999 patients, we predicted cardiovascular risk factors not previously thought to be present or quantifiable in retinal images, such as age (mean absolute error within 3.26 years), gender (area under the receiver operating characteristic curve (AUC)?=?0.97), smoking status (AUC?=?0.71), systolic blood pressure (mean absolute error within 11.23?mmHg) and major adverse cardiac events (AUC?=?0.70). We also show that the trained deep-learning models used anatomical features, such as the optic disc or blood vessels, to generate each prediction.
‘a next-generation, no-compromise automation system’.
Uses: Web-scale configuration management of all Linux/Unix systems; Application deployment; Immutable systems build definition; Maintaining stateful services such as database and messaging platforms; Automating one-off tasks & processes; Deployment and management of the undercloud. Features: Python 3 DSL; Declarative resource model with imperative capabilities; Type / Provider plugin seperation; Implicit ordering (with handler notification); Formalized “Plan” vs “Apply” evaluation stages; Early validation prior to runtime; Programatically scoped variables; Strong object-orientation
Later JDK versions have made it far easier to run a JVM application in a Linux container. The memory support means that if you relied on JVM ergonomics before than you can do the same inside a container where as previously you had to override all memory related settings. The CPU support for containers needs to be carefully evaluated for your application and environment. If you’ve previously set low cpu_shares in environments like Kubernetes to increase utilisation while relying on using up unused cycles then you might get a shock.