Couple more questions: - Is there any way to know which partitions in which caches are currently unavailable? - How can I check which nodes failed to trigger the data loss?
The only way I see is to listen for EVT_CACHE_REBALANCE_PART_DATA_LOST event... Probably we should add something more obvious to the API? -Val On Tue, Dec 19, 2017 at 11:41 AM, Denis Magda <dma...@apache.org> wrote: > Alex K., thanks for stepping in. > > It will be vivid what to do on the tooling side once it’s clear how to use > the feature in general. > > Alex G., Sam, Yakov, could you comment on this point below? > > >> And finally (most importantly), I'm a bit confused > >> by Ignite#resetLostPartitions method itself. What are the best practices > >> for using it? How a user should decide that partitions are actually > >> restored and how should he choose when to call this method? For > example, if > >> we have persistence enabled, is it enough to just bring back the nodes > or > >> something else is needed? Actually, why don't we detect this > automatically > >> in this scenario? > > — > Denis > > > On Dec 19, 2017, at 2:01 AM, Alexey Kuznetsov <akuznet...@apache.org> > wrote: > > > > Val, > > > > I'm not an expert in how caches subsystem works internally, > > but I think it is a good idea to add this info and tooling to Web > Console > > / Visor. > > > > Could you create issues with description on what part of Web Console / > > Visor it could be added. > > I guess for Visor CMD it could be some mode for "cache" command I > think. > > > > > > On Fri, Dec 15, 2017 at 8:54 AM, Valentin Kulichenko < > > valentin.kuliche...@gmail.com> wrote: > > > >> Folks, > >> > >> Since 2.0 we have introduced PartitionLossPolicy which blocks access to > >> cache if data loss occurred. This is an awesome feature, however it is > not > >> very clear how to use it properly. > >> > >> First of all, there is no documentation. Ticket already exists though > and > >> hopefully it will be completed soon: > >> https://issues.apache.org/jira/browse/IGNITE-6994 > >> > >> Second of all, looks like there is no required tooling. Visor and Web > >> Console should be able to show the status (i.e. which partitions are > >> available and which are not), fire alerts in case of partition loss, > >> provide an ability to restore partitions via Ignite#resetLostPartitions > >> method, etc. > >> > >> And finally (most importantly), I'm a bit confused > >> by Ignite#resetLostPartitions method itself. What are the best practices > >> for using it? How a user should decide that partitions are actually > >> restored and how should he choose when to call this method? For > example, if > >> we have persistence enabled, is it enough to just bring back the nodes > or > >> something else is needed? Actually, why don't we detect this > automatically > >> in this scenario? > >> > >> I would appreciate any inputs and thoughts on this topic. > >> > >> -val > >> > > > > > > > > -- > > Alexey Kuznetsov > >