jenkins test api

@NitzanMordhai What's wrong with invoking erase on the iterator? Invoking erase on the iterator should have the same effect as invoking it on the key, but much more efficiently (constant time rather than logarithmic). I don't see a mechanism that would allow this to fix the bug linked in the ticket.

@NitzanMordhai What's wrong with invoking erase on the iterator? Invoking erase on the iterator should have the same effect as invoking it on the key, but much more efficiently (constant time rather than logarithmic). I don't see a mechanism that would allow this to fix the bug linked in the ticket.

seeing that. will rework that.

@athanatos it looks like the peering process didn't call start_peering_interval which is supposed to call on_change which triggers the cleanup and cancel of the gather. we also have finish methode in the gather (struct C_gather) that getting r=-125) but doesn't clear the cls_gather_ops map, that can solve the issue as well

@athanatos Since we are not have new interval, the cancel was not invoked, i added logic to cancel the ops if we already have them in the map (just like copy ops)

jenkins test make check

@athanatos it looks like the peering process didn't call start_peering_interval which is supposed to call on_change which triggers the cleanup and cancel of the gather. we also have finish methode in the gather (struct C_gather) that getting r=-125) but doesn't clear the cls_gather_ops map, that can solve the issue as well

If peering really skipped start_peering_interval, that would be an incredibly huge bug. Do you have logs?

If peering really skipped start_peering_interval, that would be an incredibly huge bug. Do you have logs?

@athanatos i was able to recreate the error with my local build, can you check the logs on folio03? /home/nmordech/output_gather

or should i move them to somewhere else?

@NitzanMordhai See my comment https://tracker.ceph.com/issues/64258 -- there was no interval change. The entry is there because the op was retried and the previous attempt was still in progress. This feature was implemented as an experiment, but I'm not aware of any serious users. For now, I suggest:

1. move this test into an experimental suite and don't consider it for normal testing
2. mark cls_gather deprecated in squid. If someone steps up to finish the implementation and maintain it going foward we can un-deprecate it. Otherwise, we'll remove it in T.

@NitzanMordhai Can you take care of the above?

@NitzanMordhai Can you take care of the above?

@athanatos thanks for reviewing it, will do that

Just to the record:

$ grep -r cls_cxx_gather 
crimson/osd/objclass.cc:int cls_cxx_gather(cls_method_context_t hctx, const std::set<std::string> &src_objs, const std::string& pool,
Binary file crimson/osd/.objclass.cc.swp matches
objclass/objclass.h:extern int cls_cxx_gather(cls_method_context_t hctx, const std::set<std::string> &src_objs, const std::string& pool,
cls/test_remote_reads/cls_test_remote_reads.cc:    r = cls_cxx_gather(hctx, src_objects, pool, cls.c_str(), method.c_str(), *in);
osd/objclass.cc:int cls_cxx_gather(cls_method_context_t hctx, const std::set<std::string> &src_objs, const std::string& pool

$ grep -r cls_cxx_get_gathered_data ./
./crimson/osd/objclass.cc:int cls_cxx_get_gathered_data(cls_method_context_t hctx, std::map<std::string, bufferlist> *results)
Binary file ./crimson/osd/.objclass.cc.swp matches
./objclass/objclass.h:extern int cls_cxx_get_gathered_data(cls_method_context_t hctx, std::map<std::string, bufferlist> *results);
./cls/test_remote_reads/cls_test_remote_reads.cc:  int r = cls_cxx_get_gathered_data(hctx, &src_obj_buffs);
./osd/objclass.cc:int cls_cxx_get_gathered_data(cls_method_context_t hctx, std::map<std::string, bufferlist> *results)