Monitoring #44
Monitoring #44
Conversation
The value determines how many vhosts the nginx process can have, so it's important it's not too small. By default, nginx sets this to a value that depends on the CPU's cache line width. We don't want to be too sure that all solarnet machines have the same CPU, so we make the value the same everywhere. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
The role is to make metrics accessible, and Prometheus is just an implementation detail. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
This is one more step in decoupling nginx from the IPFS gateway. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
... in order to make space at the top-level for Grafana. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
Changes to the Dockerfile often require a rebuild of the image.
We can't delete the image to trigger a rebuild, because we don't
wanna kill off the Prometheus daemon.
We introduce a ref-file, similar to how cjdns upgrades are handled.
The ref-file contains the currently running git ref (commit, branch,
or tag). We build unless the file exists and contains the expected ref.
In order to trigger a rebuild, simply remove the ref-file:
ssh root@host rm /opt/prometheus.ref
License: MIT
Signed-off-by: Lars Gierth <larsg@systemli.org>
License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
Prometheus' node_exporter is a little daemon which exposes metrics for things like CPU, RAM, network, filesystems. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
Allows anyone with a password to peer with the Hyperboria network, via the solarnet hosts. Peering is needed in order to access monitoring. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
We don't want the playbook to break everytime someone fat-fingers file changes into the build directory. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
This bug is really silly: the copy step grabs the file from the *local* disk, not the actual host's disk. That worked purely out of incident. It obviously never installed the correct version of cjdns. We now install the correct cjdroute, as well as the correct upstart script. We also restart cjdroute in case the upstart script changed. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
This latest version introduces a default password for the admin API. Additional configuration for the tools is no longer neccessary. This default password is "NONE" by convention, and the tools are assume this. License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
| Communications: | ||
| - Github repos | ||
| - IRC Channel | ||
| - Mailing Lists |
There was a problem hiding this comment.
i want to keep (and expand) reference to these things here. the IPFS infrastructure is more than just the programs we run. IRC, mailing list, github, etc are all important to account.
There was a problem hiding this comment.
My idea here was to avoid duplication and keep that in ipfs/community.git, so it'd be about "using the infrastructure", while ipfs/infrastructure.git would be about "running the infrastructure".
But maybe this distinction doesn't make sense? We should maybe look into merging the two then.
There was a problem hiding this comment.
it'd be about "using the infrastructure", while ipfs/infrastructure.git would be about "running the infrastructure".
Yeah this is the distinction i was going for too. i'm +1 to moving the listing over for users to consume. still want one listing somewhere that whoever will be taking care of infra can know where everything is + make sure it's running / accounted for. (e.g. i dont expect freenode or google groups or https://botbot.me/freenode/ipfs/ to fail at all, but should still be accounted for somewhere in case someone needs to replace a piece).
|
|
|
@lgierth i think re: #44 (comment) -- maybe having just one file in /infrastructure that tells someone all the pieces they have to worry about (if we all disappear) would be useful. (/community can be the user-friendly UX to it, which may not list everything, or whatever) |
SGTM, that makes a lot of sense -- this is still infrastructure that needs to be "run" in a wider sense. |
ghost
force-pushed
the
monitoring
branch
2 times, most recently
from
7674b95
to
a9cef2c
Compare
License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
1. indentation matters 2. last variable definition wins 3. name the password list item License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
License: MIT Signed-off-by: Lars Gierth <larsg@systemli.org>
ghost
mentioned this pull request
This turned out a bit big and monolithic. The changes are contained in individual commits though, with proper description.
Monitoring is now available at http://metrics.i.ipfs.io, once you've peered with cjdns and added yourself to the whitelist of allowed clients. More information about this comes with the updated README.md, and https://github.com/protocol/infrastructure/pull/73