diff options
| -rw-r--r-- | scripts/README | 73 |
1 files changed, 73 insertions, 0 deletions
diff --git a/scripts/README b/scripts/README new file mode 100644 index 00000000..f0eb191e --- /dev/null +++ b/scripts/README @@ -0,0 +1,73 @@ +$Id$ + + + +At this point it's starting to look like the pdu objects in +rpki.left_right are a subclass of a more general class of objects that +live in SQL (duh). This makes sense as left-right protocol was +designed to let us frob a subset of the rpki database on the wire. + +Hmm, it's more complicated than that. Most of left_right maps well, +but portions (eg, <list_resources/>) are just commands and do not. +Mixin? We have a varient of this same problem with the mixins I'm +already using in irbe-cli.py. + +Also need to think about up-down protocol, currently separate but some +of these objects might map there too. This is why I originally +intended to code all of these as separate object hierarchies even +though in some cases they look awfully similar. Hmm, implement as +separate hierarchies with sideways conversion methods? Still feels +wrong, but so does turning this into one big interconnected mess. + +Ok, up-down XML is all structured as commands, data is all attributes +and (occasionally) element content, so not very close match to db. + +So this appears to break out fairly cleanly: objects shown in blue in +the repository-engine-objects.pdf picture are both left-right PDU +objects and database objects, objects shown in green are only database +objects, and left-right PDU objects not shown in that picture are just +left-right PDUs. Or something close to this, details need checking. +So this does look like some kind of mixin would be appropriate. + + + +Code to pull in an entire self_id context at once turned out to be +much too complicated. Replacement is more of a demand-paged design. +Need to be careful about making sure we save out all changes between +events, objects have a .sql_dirty attribute but I should also add a +Python set object into which we can insert any object that needs to be +saved when the current event exits. + +Haven't done anything about db.commit() and db.rollback() yet, for +that matter haven't yet whacked MySQL to enable those features. + +Access to object data attributes really ought to be through accessor +methods so that the .set() method can set the sql_dirty flag +automagically. Not done yet. + +Hmm, do we really need .sql_dirty at all? Maybe just inserting the +object into the dirty set (above) would suffice? One less thing to +screw up. + +RelaxNG schemas probably ought to be internal rather than files with +magic names or config variables pointing at files. Makefile hack to +put a Python wrapper around a .rng file, and automatically convert +.rng to lxml internal form while we're at it? + +Have started on RPKI-specific exception hierarchy (well, it might be +flat) and conversion of inappropriate uses of assert and standard +exceptions to use rpki-specific exceptions. Top level of protocol +handlers needs to include exception catches that translate these +exceptions into appropriate protocol error messages. + +Generalize serve_copy_hook() into serve_action_hook() and use it to +handle actions signaled by the various left-right boolean controls. +One fun test case for this should be bsc keypair generation. + +Some of the defined actions may be things we can't implement yet due +to lack of a real timer system. Eg, the various "do this right now" +actions are probably best handled as triggers, a la interrupt +handlers. Well, maybe. IRBE folks might prefer a version that +blocked response to the request until the whole cycle is done, once we +have an event system that can treat the query and response phases of +an rpc operation as two distinct events. |