SRE thinking: clearly, we've got a complex system-failure here. New automation has been introduced, but the feedback loops haven't been debugged yet so we're getting increased rates of more-severe-than-usual incidents.
The log system we built was so good, we managed to convince the much larger company who bought us out to maybe do that too (massively scaled up Loki for them). Mwahahaha.
One of the great joys of my career was building a telemetry system so good that people went to logs as soon as a chart-squiggle looked wrong. There was so little friction to look at prod logs people actually did it.
The longer we go from the trauma, any appetite for the progressive agenda in the Federal context fades. They don't have it yet for state/local, and what they have will fade faster than support for the fed-level.
The Winter War in Minneapolis/St. Paul turned a bunch these folks into Abolish Ice supporters, but they haven't gotten to the point of extending the concept beyond the agency that occupied and brutalized a major metro area. They're not turning around and voting for the full progressive agenda.
Camp-sweeps but also better social services.
Households like that are not emotionally or intellectually prepared for "abolish ICE" because they've fought literal decades against "loony lefties" trying that for municipal police.
They're the ones we need to win over. Somehow.
Households like that believe incrementalism is the only way to make steady progress, big leaps lead to chaos and worse outcomes. Add body-cams to police. Train the police better. Add civilian oversight. More supervised-release programs.
I grew up in a house that scorned the progressive left as "loony lefties," so named because they were all ideas and no practicalities. "We believe in good government," was the phrase of my childhood. This household (which I haven't live in for decades) was shaking their head at defund the police.
Unlike Walmart, which is known for the best prices, or Amazon, which has stood out with convenience, Target wants to create a fun, treasure hunt atmosphere in stores where shoppers come in to find what’s new.
< All this: BUT NO APOLOGY TO BLACK FOLKS! 😏
I believe I've managed to almost entirely degoogle my email usage.
The SF Fed found that Treasury yields now respond 3x more strongly to oil supply news than they did pre-2021.
I think this matters to #econsky because of what central banks are staring at right now, because of what is going on, with Trump's attacks on Iran.
www.governance.fyi/p/wall-stree...
And, all the way back last March, I wrote about the capitulation of Paul Weiss and how it was “dangerous for the rule of law and the nation.”
Thankfully, others — Perkins Coie, Jenner & Block, WilmerHale, and Susman Godfrey — did fight back.
In the end, the Platform Engineer (staff+ though they may be) is merely one voice in the room. If Data Engineers are also in the room when this comes up, general purpose data-lake tooling becomes a lot more attractive *to the room*.
When on-prem becomes a thing, your highest probability pipeline will be companies already using InfluxDB-cloud as that reduces the delta-change in their telemetry production and shipping pipelines. Influx pricing almost definitely has this use-case in mind already.
One company I worked for would prioritize time-to-productivity first and ended up with several discrete database types as a result, with their own teams to support them. Another company preferred to single-stack where possible and would throw engineers to write glue code.
As an engineer, I'd prefer the purpose built metrics database every time. Problem is, that kind of decision is typically made at the director level where the balancing point is time-to-productivity vs engineering resources to write glue code.
Convergence between Engineering and Business telemetry systems has been ongoing for the last 5ish years and will continue.
The solution here depends on whether the Business Intelligence and Engineering Telemetry teams talk to each other. If they're in radically different orgs, InfluxDB v3 has a real chance. If they cooperate a lot, metrics could get shoved into Dbricks.
I can make a real case. The tricky part is convincing management not to use Databricks for metrics. Databricks is already using Parquet, we have whole departments dedicated to maintaining our Dbricks pipelines, why should we onboard another Parquet system?
The differentiator is generally price for SaaS-model databases like that. Parquet can scale extremely far, and businesses like Databricks are built on top of exactly that.
The real question is "when business needs prompt insourcing the metrics database, would I pick influxdb v3?"
Would I recommend InfluxDB v3 to a startup greenfielding their own metrics stack? Debatable. Startups generally avoid self-hosting, and that's what I did in this thread. They'd go for a cloud-based metrics provider that met their economic needs. v3-cloud would perform on par with others.
In the end my tiny home metrics database is now a lot more future-proof, which is really want I wanted. InfluxDB v3 fixed a lot of issues from their v1, which had significant business impacts for them as they misjudged the size of "small, constrained by cardinality" databases in the market.
It also took VASTLY more disk-space. The v1 database was maybe 6GB all told. The v3 data directory is 19GB. The influxdb process also takes a lot more memory. These attributes are extremely familiar if you've dealt with migrations to parquet from anything else. The price of high cardinality support.
v3-enterprise is free for home use, but is limited to a single node and only two execution cores. I'm not doing deep crunching, this is tiny home stuff. V3-enterprise creates summary tables in parquet to support much longer query windows. Once everything got done chewing, I had year+ again.
V3-core was fine for system data but had a 14 day query window limit due to how the parquet tables were laid out. That wasn't going to work for my toy power-grid metrics gatherer where I regularly look at 90+ day data, and metrics are only generated once an hour.
Happily, they have home use in mind
I noticed v3 came out and went Parquet like the rest of the industry. That was Interesting, since Parquet is the de facto solution to high cardinality data. They also put the old query language back in while also supporting SQL-like language. So I gave it a try.
I never bothered with v2 because it didn't solve the core cardinality issues, and forced a query-language change I didn't want to deal with. This was a toy at home, not anything supporting revenue. I could stay old the old, working stuff.
Two weekends ago I updated the tiny metrics database at home from @influxdb.com v1 to v3. This thing tracks system metrics from two systems and is the back end for a toy local power-grid metrics gatherer I wrote. The system metrics don't need to be kept for long periods, but power needed year+.
If you were OK with “Substack, the Nazi Bar,” maybe “Substack, a front for the gambling on life and death casino” might convince you.
"The United States suffered its first casualties on Sunday, while only 27% of voters support the offensive" says the headline.
27% eh? I guess only the Trump ride-or-die segment are behind this.
