On Thu, Aug 8, 2019 at 11:49 AM Jean-Marc Spaggiari <jean-m...@spaggiari.org> wrote:
> Don't we first write to WAL and then memstore? Getting confused here, > sorry. > > You are right JMS. IIRC, Udai is correct for older versions of HBase; the memstore update would get rolled back if the concurrent WAL write failed. S > JMS > > Le mer. 7 août 2019 13 h 22, Udai Bhan Kashyap (BLOOMBERG/ PRINCETON) < > ukashy...@bloomberg.net> a écrit : > > > memstore is written first and if write(s) to WAL fails, they are rolled > > back. > > > > From: dev@hbase.apache.org At: 08/07/19 13:15:27To: > dev@hbase.apache.org > > Subject: Re: does failure of write to memstore has any impact on response > > code from hbase > > > > On Mon, Aug 5, 2019 at 9:36 PM Maneesh Bhunwal < > maneesh.bhun...@gmail.com> > > wrote: > > > > > Hi Team, > > > > > > First fo all thanks for the awesome product. > > > > > > Can you please help me with how will application behave when write to > > > memstore fails but write to WAL has succeeded already,will we return > > > success to the user or failure? > > > > > > Failure. Client will get an exception that varies dependent on failure > > type. > > > > > > > Also what if db crashes after writing to WAL, when db will come up > after > > > crash, it will assume that whatever is there in WAL has been > replicated, > > > but that may not be the case. > > > > > > > > DB does not make progress until the write to the DB has been sync'd > (which > > means the edit has been replicated). So on crash, the WAL will be > replayed. > > The edit that was written to the WAL but not to the memstore on which the > > client received an exception, will perpetuate. > > > > S > > > > > > > Can you please help me with the above? > > > > > > Regards > > > Maneesh Bhunwal > > > > > > > > > >