s3 faq mentions "amazon s3 buckets in regions provide read-after-write consistency puts of new objects , eventual consistency overwrite puts , deletes." however, don't know how long takes eventual consistency. tried search couldn't find answer in s3 documentation.
situation:
we have website consists of 7 steps. when user clicks on save in each step, want save json document (contains information of 7 steps) amazon s3. plan to:
- create single s3 bucket store json documents.
- when user saves step 1 create new item in s3.
- when user saves step 2-7 override existing item.
- after user saves step , refresh page, should able see information saved. i.e. want make sure read after write.
the full json document (all 7 steps completed) around 20 kb. after users clicked on save button can freeze page time , cannot make other changes until save finished.
question:
- how long take aws s3 save , load item? (we can freeze our website when document being saved s3)
- is there function calculate save/load time based on item size?
- is save/load time gonna different if choose s3 region? if best region seattle?
i wanted add @error2007s answers.
how long take aws s3 save , load item? (we can freeze our website when document being saved s3)
it's not not find exact time anywhere - there's no such thing exact time. that's "eventual consistency" about: consistency achieved eventually. can't know when.
if gave upper bound how long system take achieve consistency, wouldn't call "eventually consistent" anymore. "consistent within x amount of time".
the problem becomes, "how deal eventual consistency?" (instead of trying "beat it")
to find answer question, need first understand kind of consistency need, , how eventual consistency of s3 affect workflow.
based on description, understand write total of 7 times s3, once each step have. first write, correctly cited faqs, strong consistency reads after that. subsequent writes (which "replacing" original object), might observe eventual consistency - is, if try read overwritten object, might recent version, or might older version. referred "eventual consistency" on s3 in scenario.
a few alternatives consider:
don't write s3 on every single step; instead, keep data each step on client side, , write 1 single object s3 after 7th step. way, there's 1 write, no "overwrites", no "eventual consistency". might or might not possible specific scenario, need evaluate that.
alternatively, write s3 objects different names each step. e.g., like: after step 1, save
bruno-preferences-step-1.json; then, after step 2, save resultsbruno-preferences-step-2.json; , on, save final preferences filebruno-preferences.json, or maybebruno-preferences-step-7.json, giving flexibility add more steps in future. note idea here avoid overwrites, cause eventual consistency issues. using approach, write new objects, never overwrite them.finally, might want consider amazon dynamodb. it's nosql database, can securely connect directly browser or server. provides replication, automatic scaling, load distribution (just s3). , have option tell dynamodb want perform strongly consistent reads (the default consistent reads; have change parameter consistent reads). dynamodb typically used "small" records, 20kb within range -- maximum size of record 400kb of today. might want check out: dynamodb faqs: consistency model of amazon dynamodb?
