java - Storing a list of objects on appengine across all instances -

i have app on google app engine. processes realtime data without regard past data(so no datastore/full persistency), , don't need extremely reliable recover data in case of instance crash or new deployment.

i have multiple instances, due load, need share single list of objects. looked memcached, requires mapping key object in map-style interface without ability dump entries.

in addition, i'm hoping make entries expirable unless "bumped", though can handle in user code.

what's reasonable solution doesn't involve datastore? i'd prefer able handle concurrent reads , writes, multiple app engine instances(which why simple java object declaration doesn't work it's per-instance).

memcached potentially store single object require getting object, getting sort of lock, modifying it, , writing cache.

n.b. there unrelated data in presistent app engine datastore.

clarification: need all entities. entities may modified @ time, removed expiration or semi-automatically, or created scratch. i'm not particularly able head billable backends.

edit: yes, work single lists stored locally expect spikes multiple instances needed. i'd need @ least best-effort approach getting full list, though don't mind delay of synchronization. however, round-robining communication between instances sans backend relatively unscalable, , i'd stuck in more of rent-a-server scenario there.

i don't care if changes don't make through after change committed on frontend instance. changes need make through 100% certainty, however. see insertions , updates lastseen value(long) increased. updates should replicated or reflected on instances.

note: bounty got refunded due fact unable award due circumstances @ time, cannot control. re-add when possible.

what's wrong datastore here?

if concern you'll rack expense having lot of items in db, save each item datetime_created field auto_now_add=true (or java equivalent; come python), , schedule cron go through , delete old?

the datastore supposed highly available concurrent reads , writes, far i'm aware.

you might consider moving off appengine. redis sounds right tool job you're describing, , elastic beanstalk on aws handle needs quite there. of course, it'd more effort things set up, not overwhelmingly so.