NB: This is not a new issue (it existed in 20.1) however it just caused me to spend a lot of time staring at my screen not understanding what was going on.
What I did:
- on node n1,
cockroach start without --join (or a self-join)
- on node n2,
cockroach start --join=n1
At this point I verified in logs that n2 was establishing a conn to n1, and n1 was receiving an init request / join request from n2.
- then I ran
cockroach init --host=n2
That caused n2 to initialize a cluster centered on n2 (with node ID 1). However n1 does not participate and does not join, and remains uninitialized (no node ID).
This surprised me -- I thought that the join flag need not create a full connectivity graph for the init command to propagate throughout the established connections.
Was that the intended behavior?
cc @tbg @bdarnell @irfansharif
NB: This is not a new issue (it existed in 20.1) however it just caused me to spend a lot of time staring at my screen not understanding what was going on.
What I did:
cockroach startwithout--join(or a self-join)cockroach start --join=n1At this point I verified in logs that n2 was establishing a conn to n1, and n1 was receiving an init request / join request from n2.
cockroach init --host=n2That caused n2 to initialize a cluster centered on n2 (with node ID 1). However n1 does not participate and does not join, and remains uninitialized (no node ID).
This surprised me -- I thought that the join flag need not create a full connectivity graph for the init command to propagate throughout the established connections.
Was that the intended behavior?
cc @tbg @bdarnell @irfansharif