Report from the DataStore Design Team Breakout Session draft-nmdsdt-netmod-revised-datastores-00 IETF97
Breakout Session Meet Wednesday 1:30-7pm Lots of questions and clarifications No changes in the design A positive experience Walked thru Implications and Issues
New Issues in the draft Examples More descriptive text Timeliness Data in <running> should appear in sync with <intended> Data in <applied> should appear in sync with <operational-state> "Factory default" config
Example 1: simple config Straight-forward config example Full walk through edit-config->running->intended->applied->op Discuss templates and inactives
Example 2: system-controlled Loopback interface implicitly created by system In <operational-state> <interface origin="system><name>lo0<...> Part 2 adds user config Now explicitly in <running> and <intended> <interface origin="static"><name>lo0<...>
Example 3: delayed cleanup BGP Peer list Explicitly created in <running> Fully populated In <operational-state> with origin="system" local-address and local-port Delete five peers and commit Peers continue to appear until "released" Sockets closed and local resources released In both <applied> and <operational-state>
Example 4: ephemeral data Interfaces created in <running> No installed FRU Interfaces do not appear in <applied> or <operational-state> When FRU is plugged in: Interface is created and populated with data Unconfigured ports of installed FRUs will exist Statistics are initialized to zeroes (like new) Key is that FRUs do not break config validation Only FRU-related, not "logical" interfaces LAG interface with no installed member interfaces
Example 5: dynamic data Unnamed dynamic data source Don't want to address the i2rs issues here Data merged with <intended> into <active> Issue: fix arrows for "after" control-plane datastores?
Issues <active> No need constraints on <applied> and <operational-state> SHOULD for "when" "expected" for others No point modeling rules that aren't followed, but Postel Principle applies <applied> in RESTCONF: take to NETCONF WG Disagreement of goals of RESTCONF Does it need all NETCONF capabilities? Is it the "Easy" button?