Skip to content

Conversation

@andreaspeters
Copy link
Member

No description provided.

urbanserj and others added 30 commits August 12, 2019 15:39
IPSet Netlink Protocol splits big sets into several netlink messages.
First message contains all metadata about a set, but just a part of
all entries. Following messages contains only remaining entries.
Add support for big sets in the ipset mgr.
Move from Lager to Logger
Leverage exponential backoff with requests to Mesos
Add syntax_tools, compiler, and backoff as runtime deps
Mesos listener is honest now again when asked if it is a leader or not
Wait till agents become active before fanning out Mesos tasks
Task update leads to two DNS zone updates
DC/OS overlay networks should be compared by-value
Remove support for set mode from dcos dns
Move common logic from dcos_net_mesos to new behavior, so
it can be re-used in l4lb later.
L4lb manager receives update from mesos leader. Load balancer
doesn't have to check reachability of nodes because it always
have consistent state.
urbanserj and others added 29 commits November 8, 2019 13:36
Operator Streaming API in L4LB
Codified networking run book
Also,
 - make the script compatible with DC/OS >= 2.0
 - fix indentation
 - make it use jq only if it is available
Wrap the "main" logic into a `main` function that allows for easier use
via `curl <url> | bash` should a user desire to do so.
This reverts commit c0f07b2, reversing
changes made to eb72421.
Revert "Merge pull request #189 from dcos/urbanserj/l4lb"
Make runbook a bit more sysadmin friendly
Instead of staticly setting the weight to `1` always, set it only to `1`
when in a `running` state. `killing` and any other state will be set to
`0` to prevent new connections from being sent to the backend, but still
allow existing connections through.
When a task is unhealthy, we should not sever conenctions by removing it
from the VIP. Instead set the weight to 0 to prevent new connections,
while allowing existing connections.
Handle TASK_KILLING state for loadbalancing
Adds testing that the different backend selection and manupulation
functions work with and without the weight being specified. Confirms
that if an agent populates a VIP using the old code that the new code
won't crash.
Since the network runbook can be used as a lite "bundle" capture more
data.
runbook: capture vxlan state and nclock
@andreaspeters andreaspeters merged commit 0db6fde into AVENTER-UG:master Feb 11, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants