Re: [jira] Commented: (HBASE-1249) Rearchitecting of server, client, API, key format, etc for 0.20

[Erik Holstad <er...@gmail.com>]

The reason that I thought that putting the newest version of the flush as
the "id", was because when we let the client put whatever timestamp in the
storefiles,
"now" might not be related to them.

RE: [jira] Commented: (HBASE-1249) Rearchitecting of server, client, API, key format, etc for 0.20

[Jonathan Gray <jl...@streamy.com>]

Understood.

I think timestamps should be considered and thought about strictly as
datetime stamps.

The setting of past stamps should be permitted to allow users to insert
things with the actual stamps (of a crawl, for example) rather than insert
time only.

I don't see a downside of using the newest version stamp in the flushed
store, except that you have to determine it and that requires a scan.  And I
believe we need this seqid prior to actually performing the flush so you'd
have to scan memcache first.

Also, this gives you an additional point of information about the time of
the flush.  The newest version in a storefile can always be determined, if
necessary, from data.

JG

> -----Original Message-----
> From: Erik Holstad [mailto:erikholstad@gmail.com]
> Sent: Wednesday, April 01, 2009 12:59 PM
> To: hbase-dev@hadoop.apache.org
> Subject: Re: [jira] Commented: (HBASE-1249) Rearchitecting of server,
> client, API, key format, etc for 0.20
> 
> The reason that I thought that putting the newest version of the flush
> as
> the "id", was because when we let the client put whatever timestamp in
> the
> storefiles,
> "now" might not be related to them.