Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by sofixa on Wednesday June 23, 2021 @06:16AM
(#27602797)
I feel like InfluxDB were the best time series database in town for some time (2015-2018?), but they failed to monetise and exploit the space properly, and now they have lots of competition, many of it better in a lot of cases ( e.g. speed, space use, ease of use, pricing, integrations), be it for self-hosted or cloud.
Some of it was wasted effort ( Cheonograf is still useless compared to Grafana, who thought it's a good idea to compete with it?), some was poor luck, some was decisions that backfired ( like removing clustering from the free version). Some of it was everyone deciding pull is better than push ( i disagree in most cases, but see the point).
Telegraf still remains a very solid collecting agent and IMHO the best in class. With everyone (from the SaaS monitoring vendors ) moving to OpenTelemetry i think that in some time it'll fall behind as well.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by karterk on Wednesday June 23, 2021 @07:28AM
(#27603223)
InfluxDB could have become Grafana if they had focussed better. It was a breathe of fresh in a space where existing solutions were horrendously complicated. I suspect that they were under a lot of pressure to monetize and lost some way in attempting to do that.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by heipei on Wednesday June 23, 2021 @07:57AM
(#27603400)
Telegraf is the main reason I'm still running the Telegraf + InfluxDB + Grafana stack. I might have checked out TimescaleDB or others, but since Telegraf works great with InfluxDB and has all sorts of arcane collection plugins, mostly with sensible options, I'm staying put. Good luck replicating the years of community effort to build out all these collectors to anyone who wants to replace it.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by KaiserPro on Wednesday June 23, 2021 @06:22AM
(#27602826)
Telegraf's USP is that its an anything to anything converter. Which is invaluable when you are supporting a whole bunch of mixed stuff.
I'm not sure OpenTelemtry is a replacement for actual designed metrics though. Its great for getting info about the state of an instance of something. but thats only half the picture. Building meaningful alerts of just raw telemetry is hard.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by KaiserPro on Wednesday June 23, 2021 @06:17AM
(#27602803)
Old man opinion here:
I really like grafana, personally I'd probably host it myself even now. (At all previous companies I did.) Its well contained and even with alerting its simple to set up and sync state (assuming you are using a hosted database, which can be expensive)
Where my opinion gets controversial: I dislike Prometheus, and I really don't like influx. I much prefer graphite.
Influx has higher temporal resolution, and an SQL interface (its the SQL interface I don't like)
Prometheus has a pull model, which I can see the appeal of, but I don't really like. But it deals with ephemeral metrics much better (however you need a gateway to cache the metrics, as they might not get pulled in time.)
The issue is that graphite's interface is much better for discovering and filtering/transforming your data. I'm not the worlds greatest SQL person, so I can't do that much complex stuff, I also don't know how to do maths in SQL well. With graphite I don't need to worry, all the functions are there in menu, derivative? yes, all the averaging? mostDeviant? yes.
Graphite made making graphs quick and easy, I don't think I can say the same for influx or prometheus. I know that all sorts of non technical people can use graphite/grafana. I'm less convinced that they can use Prometheus or influx as well without training.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by wperron on Wednesday June 23, 2021 @07:26AM
(#27603206)
> Prometheus has a pull model, which I can see the appeal of, but I don't really like.
That's such an antiquated statement at this point. Prometheus might have been pull-only in a distant past, but the remote write protocol has been alive and well for a long time now, and it works amazingly well, at surprising scale.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by heipei on Wednesday June 23, 2021 @07:55AM
(#27603381)
I use Grafana religiously every day, rely on it for metrics, long-term trends and alerting. I also use Telegraf with InfluxDB, but thanks to Grafana I've never had to write a single InfluxDB statement.
Also: It's not just higher temporal resolution, it's having an actual time-series database vs a metrics database. With InfluxDB (and all other event-based databases) I can go in after the fact and just plot the CPU load for one specific host, or a group of hosts, or plot the HTTP requests that took longer than 200ms, hit a specific endpoint and finished with a specific HTTP status code. If you don't keep the raw measurements you always have to make decisions about which groups you aggregate down to ahead of time. Just something to keep in mind.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by pqvst on Wednesday June 23, 2021 @06:28AM
(#27602864)
When we were self-hosting Grafana with an influx data store writing queries was actually very easy since Grafana provides a "query editor" to generate the SQL for Influx. Ironically that was much easier to use than InfluxDB Cloud's own solution (which now uses a completely different query language).
Grafana Cloud seems to mostly advertise the pull model, whereas we were really looking for a push model. I'm glad I figured out how to get that to work with Graphite, since Grafana Cloud offers both Prometheus and Graphite.
Now given that Grafana Cloud's free-tier covers our needs I'm glad we no longer have to maintain a self-hosted version (and we're definitely getting better performance now).
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by KaiserPro on Wednesday June 23, 2021 @06:31AM
(#27602878)
I should have written that I fully understand why you'd want to sling it out and be hosted by someone else.
I'm a devop/sre/pe/sysadmin so I have a motivation to keep metrics in my firm grasp. Given the price point and performance, for small organizations it'd be madness to Not get grafana cloud to host it.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by kungito on Wednesday June 23, 2021 @08:53AM
(#27603938)
It honestly doesn't feel powerfull at all. I admit, I have started using Influx 2.0 during beta, but the experience has been horrible. I first tried to write all my logic in Influx and have now over few months moved everything back into Rust with just lines which go directly into graphs staying in influx. The language as it is is ok, it's very basic, but the runtime and the development environment is very unfinished. So many internal errors whatever you do, the responses are slow to be received because I guess so few people try to write anything more complicated in Flux, the debugging is very hard. You can't even print out single values, everything has to be displayed in graphs. They support using arrays and single values but you just get convoluted internal errors and compiler panics if you don't write correctly in a language without explicit typing.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by KaiserPro on Wednesday June 23, 2021 @06:25AM
(#27602849)
> It's weird and has a learning curve, but powerful.
I don't doubt that, but teaching it to business analysts, product owners and the like is a real headache.
I'm sure I would grow to love it, but then it would require me to make dashboards for other people. Getting them to make them themselves is living the dream
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by gcbirzan on Wednesday June 23, 2021 @08:23AM
(#27603611)
SQL is not limited, InfluxQL is. And, flux is incredibly verbose and error prone (at least a year ago when we last looked at it). However, with the transformations in grafana, there's no reason with flux, in the majority of cases.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by jstrong on Wednesday June 23, 2021 @10:27AM
(#27604923)
influxql is limited, yes, but much less verbose for time-based aggregation than the equivalent sql queries, which is why I like using it for adhoc visualization and inspectibility. (I don't like flux). to me it's great for a limited purpose. writing the equivalent sql queries would drive me up the wall.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by gcbirzan on Thursday June 24, 2021 @07:19AM
(#27616088)
For sure, we were evaluating timescaledb and influx, we quickly came to the conclusion that while a programmer could write SQL queries to do what influxql lets you do, they're very complicated compared to the builtin UI that even non tech people can understand.
Re:Hosted Monitoring: Evaluating InfluxDB Cloud and Grafana ...
by mikojan on Wednesday June 23, 2021 @12:16PM
(#27606221)
InfluxDB Cloud might be the single best solution if what you want is a managed service that just works somehow.
As a time series database however it is rather painful.
You can't have InfluxDB only. It comes integrated with the complete ecosystem. For example, it exposes an app/web server by default which you can't disable.
Flux DSL is decent but so was SQL and I already knew SQL.
Shipping a database does not appear to be topping their priority list and it's kind of irritating. A year into the InfluxDB 2 release you weren't able to delete data. Now you can, but you can't be sure when that's going to actually happen. And there's no way of monitoring/managing currently running queries. Add to that its massive memory consumption and you're left wondering what other database solution would get away with this.
Some of it was wasted effort ( Cheonograf is still useless compared to Grafana, who thought it's a good idea to compete with it?), some was poor luck, some was decisions that backfired ( like removing clustering from the free version). Some of it was everyone deciding pull is better than push ( i disagree in most cases, but see the point).
Telegraf still remains a very solid collecting agent and IMHO the best in class. With everyone (from the SaaS monitoring vendors ) moving to OpenTelemetry i think that in some time it'll fall behind as well.
I'm not sure OpenTelemtry is a replacement for actual designed metrics though. Its great for getting info about the state of an instance of something. but thats only half the picture. Building meaningful alerts of just raw telemetry is hard.
Our Graphite hosting is also best in class and this makes me happy to see the good feedback!
PS: I work on Grafana Cloud Prometheus.
I really like grafana, personally I'd probably host it myself even now. (At all previous companies I did.) Its well contained and even with alerting its simple to set up and sync state (assuming you are using a hosted database, which can be expensive)
Where my opinion gets controversial: I dislike Prometheus, and I really don't like influx. I much prefer graphite.
Influx has higher temporal resolution, and an SQL interface (its the SQL interface I don't like)
Prometheus has a pull model, which I can see the appeal of, but I don't really like. But it deals with ephemeral metrics much better (however you need a gateway to cache the metrics, as they might not get pulled in time.)
The issue is that graphite's interface is much better for discovering and filtering/transforming your data. I'm not the worlds greatest SQL person, so I can't do that much complex stuff, I also don't know how to do maths in SQL well. With graphite I don't need to worry, all the functions are there in menu, derivative? yes, all the averaging? mostDeviant? yes.
Graphite made making graphs quick and easy, I don't think I can say the same for influx or prometheus. I know that all sorts of non technical people can use graphite/grafana. I'm less convinced that they can use Prometheus or influx as well without training.
That's such an antiquated statement at this point. Prometheus might have been pull-only in a distant past, but the remote write protocol has been alive and well for a long time now, and it works amazingly well, at surprising scale.
Also: It's not just higher temporal resolution, it's having an actual time-series database vs a metrics database. With InfluxDB (and all other event-based databases) I can go in after the fact and just plot the CPU load for one specific host, or a group of hosts, or plot the HTTP requests that took longer than 200ms, hit a specific endpoint and finished with a specific HTTP status code. If you don't keep the raw measurements you always have to make decisions about which groups you aggregate down to ahead of time. Just something to keep in mind.
Grafana Cloud seems to mostly advertise the pull model, whereas we were really looking for a push model. I'm glad I figured out how to get that to work with Graphite, since Grafana Cloud offers both Prometheus and Graphite.
Now given that Grafana Cloud's free-tier covers our needs I'm glad we no longer have to maintain a self-hosted version (and we're definitely getting better performance now).
I'm a devop/sre/pe/sysadmin so I have a motivation to keep metrics in my firm grasp. Given the price point and performance, for small organizations it'd be madness to Not get grafana cloud to host it.
SQL was too limited, that's why they did Flux, their new query language. It's weird and has a learning curve, but powerful.
I don't doubt that, but teaching it to business analysts, product owners and the like is a real headache.
I'm sure I would grow to love it, but then it would require me to make dashboards for other people. Getting them to make them themselves is living the dream
As a time series database however it is rather painful.
You can't have InfluxDB only. It comes integrated with the complete ecosystem. For example, it exposes an app/web server by default which you can't disable.
Flux DSL is decent but so was SQL and I already knew SQL.
Shipping a database does not appear to be topping their priority list and it's kind of irritating. A year into the InfluxDB 2 release you weren't able to delete data. Now you can, but you can't be sure when that's going to actually happen. And there's no way of monitoring/managing currently running queries. Add to that its massive memory consumption and you're left wondering what other database solution would get away with this.
For personal projects I have since moved to IoTDB