makeconfluentcfg exports enclosure.manager and enclosure.bay from the mpa
and id columns. Those columns belong to the mp table. When the command runs
with a node range it reads them from the mp table, but when it runs with no
argument it reads them from the nodepos table, which has only node, rack, u,
chassis, slot, room and height. The two attributes were therefore always
empty for a whole cluster export.
Read them from the mp table in both branches.
makeconfluentcfg gives confluent the console settings, the credentials of the
hardware manager, the location and the enclosure of each node. It does not
give the switch and the port that the node is cabled to, which confluent uses
to find a node by the port it answers on.
Read the switch table and give confluent net.switch and net.switchport.
A node has one row in that table for each of its interfaces, thus keep every
row and not the first one. A row that names an interface gives
net.<interface>.switch and net.<interface>.switchport, so a node with more
than one interface keeps the port of each. A row that names no interface
gives the names without an interface.
Read the table with the node list when the command receives a node range and
read the whole table when it does not, as the command already does for the
other tables.
A cluster whose switch table is empty receives the configuration that it
receives today.
Recovered from the lenovobuild branch, where this arrived as one commit and
two repairs of it: the first keeps one row for each node, which loses every
interface but one, and reads the switch columns from the nodepos table in the
branch that takes no node range, where that table has no such columns and the
feature does nothing.
preprocess_request skips any node whose nodehm entry has neither cons nor
serialport. That is the right default while scanning the whole table, but it
also drops nodes the administrator named on the command line, leaving the
dispatched node list empty. makeconfluentcfg then falls back to reading every
node, so naming one console-less node quietly configures the whole cluster
instead of the node that was asked for.
A named node with no nodehm row at all is worse: both the selection here and
the second lookup in makeconfluentcfg produce an undefined entry, which reaches
confluent as an empty node name.
Apply the skip only to the full table scan, and carry the node name through
both lookups when the nodehm row is missing. Nodes without a console method are
already handled downstream, where makeconfluentcfg keeps explicitly named
entries and donodeent falls back to hardware management credentials.
Co-authored-by: Jarrod Johnson <10814490+jjohnson42@users.noreply.github.com>
makeconfluentcfg -d erroneously would exit after only considering
the first possible node. Allow it to iterate over all the
confluent nodes seeking a match.
Vinícius Ferrão
GONG Jie
banuchka
Jarrod Johnson
Mark Gurevich
Victor Hu
Jarrod Johnson