acidmath
These are just my anecdotes, for sure, but (also anecdotal) I rarely hear of other "ops" type people using Chef, and most of the ones I know never got more than just their feet wet with Chef (the SaltStack beta was out early enough to avoid Chef).
I'm a 3/4 black partially Jewish gay man, so you can imagine where I stand on "Free Speech absolutism". While I am for the most part on your "side", I want to stick to arguments that are persuasive.
Anyone with access to NSA plus various subcontractors' toolsets can "unmask" these people in like five minutes. Musk may not be "tech genius" some of the media makes him out to be, but he knows enough about how the internet and computers work (or has advisors who do) to figure that out.
To put it bluntly, I live in the deep South in an 85% white neighborhood. Most of them voted for Trump. While I may be a mixed race (mostly black) guy who likes it in the butt, my neighbors aren't trying to lynch me. Any notion that they are is just not in touch with reality. The only people I get actual hate from are other blacks. I tell batshit liberal white women to their face I don't hate Trump and they are still nice to me 'cause they like the BBC.
While MSNBC may have told me a different story, I know the news isn't real. When I saw MSNBC pre-election I don't see "news", I saw mental illness, what it looked like was mental illness.
Anyone who considers themselves "hacker", or even just above midwit, should have figured it out a long time ago: the news isn't real.
In some earlier Ceph clusters I was responsible for I set replication to four or even five.
There was a point where I wrote an internal memo to business leaders explaining that had I not set that replication to four or five we would have been unable to meet business goals (we would have lost the contracts keeping us afloat during critical stages), but now at scale the real dollar cost of that redundancy was showing. In that memo I explained what networking we needed vs what we had, what people we needed vs what we had, and so on. I eventually got the networking gear the company needed in place and was able to hire the kind of people we needed.
For Ceph clusters I am chiefly responsible for today we have Ceph pools in each availability zone doing erasure coding (four or more parity bits) and a small service makes sure objects are copied between the distanced datacenters (availability zones).
While I do get a bit of a boner (of the Hank Hill on propane kinda vibe) on how the erasure coding is distirbuted across cabinet rows, to be fair it is a bit of an optimization.
_
> to have lower durability (say replicas are located within the same networking pod)
A network segment should never ever go down. I know in some places this is "optimistic". I can say for us it's not "optimistic" because we don't allow Ethernet protocol to "hop" anywhere. Layer-2 broadcast domains either stop at top-of-rack and the rest is layer-3 or we are using Infiniband. Stable Ethernet networks are a thing, but not using Ethernet beyond where Ethernet belongs avoids so much risk.
100% network uptime is achievable.
In some places more critical nodes are connected to three Infiniband switches with two uninterrupted power supplies dedicated to each Infiniband switch. Maybe that's "excessive", but within the last year we had to contend with a switch failure. Then a week later a datacenter provider (one that will tell you they never ever EVER lose power) for one of our availability zones was unable to provide power for over two full days.
I don't think we're ever going to have downtime or lose data. In other places this would be "optimistic", but for this deployment I have access to the resources I need to achieve the thing. The only thing that has me concerned about downtime is the current WW3 we are already in heating up.
*honk honk*
Free speech absolutism shouldn't extend to: advertising that is unsolicited and disseminated in a bad faith style.
Sorry about the state of your reading comprehension (or lack of it). Unfortunate.
When I read this I think "but you should never lose an object". Do you mean like the underlying data chunks Ceph stores? Can you elaborate on this part? I know some of the teams I work with do things in unorthodox ways and we tend to operate on different assumptions than others.
> so pg’s can be spread across racks/datacenters.
Some Ceph pools come to mind (this was a while ago, I'm sure they're still running though) where the erasure coding was done across cabinet rows and each cabinet row was on its own power distribution. I don't know how the power worked but I was told rather forwardly that some specific Ceph pools' failure domains aligned with the datacenter's failure domains.
> We already have a loki setup
Nice. We have logs go into S3 and then anyone who prefers a particular tool is welcome to load whatever sets of logs from S3 within the resource limits set for whatever K8s namespace they work with. Originally keeping logs append-only in S3 was for compliance but we wanted to limit team members by RAM quota rather than tools in line with the "people over tools over process" DevOps maxim.
Beyond that practical matter, the least bad definition I have for spam (and this is just off the top of my head right now) is: advertising that is unsolicited and disseminated in a bad faith style.
Any of us who are being honest with ourselves, maybe with exception to the acutely socially challenged, know spam when we see it.
Run Ceph on https://rook.io/ ; don't bother with Cephadm. Running Rook provides very helpful guard rails. Put the logs for Ceph Rook into Elasticsearch+Kibana on its own small (three or four node) dedicated Ceph Rook cluster. Which Kubernetes distro this runs on matters more than anything.
Recently we are looking at using https://www.parseable.com/ instead of Elasticsearch+Kibana. And we had somewhat recently started moving things from Elaticsearch+Kibana to OpenSearch+OpenSearchDashboards due to the license change.
The requirement outlined by Ceph documentation to dedicate layer-1 paths (can be same switches, but must be different ports) to Ceph replication is not about "performance" but about normal functionality.
If you have any pointed questions feel free to email "section two thirty audit@mail2tor dot com" (where "two thirty" are the three digits rather than spelled out).
In newer DDR5 servers where we can't get NVDIMM, the alternative battery backed RAM options leave us with even less to work with.
Where we have counts of HDDs or SATA/SAS SSDs in the hundreds, we still want the performance improvements provided by WAL (or functional equivalent such as ZiL/SLOG) on NVDIMM and some layer-2 (where layer-1 is RAM) caching with NVMe.
Ceph OSDs want a dedicated WAL device. Some places use OpenCAS to make "hybrid" devices out of HDDs by pairing them with SSDs where the SSDs can accelerate reads for that HDD and the Ceph OSD goes on a logical OpenCAS device. OpenCAS is really great, but the devices acting as "caching layer" often end up underutilized.
By placing "big" Ceph OSDs on ZVOLs, we don't have individual disk slices for WAL (or equivalent) or individual disks for layer-2 read caching, but a consolidated layer in the form of ZFS Intent Log on "Separate Log" (NVDIMM) and another consolidated layer in the ZFS disk pool's L2ARC (layer-2 adaptive readback cache).
The ZVOLs are striped across multiple relatively large RAIDz3 arrays. Yeah, it's "less efficient" in some ways, but the tradeoff is worth it for us.
https://docs.ceph.com/en/latest/rados/configuration/bluestore-config-ref/#devices
https://open-cas.com/Where in the US Constitution do you believe this is stipulated?
> democracy
The USA is not a democracy. The USA is a larger republic of a union of republics.
Does there exist any example of an administration (federal or state) successfully prohibiting the private sale of an LGBTQ book?
Even the poorest Americans can afford coconut oil and Fennel oil.
There's no reason to drink that shit (fluoride), there's no reason to brush with it either.