ignite-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alexey Goncharuk <alexey.goncha...@gmail.com>
Subject Re: IGNITE_HOME for persistence
Date Tue, 30 Jan 2018 20:55:07 GMT
Denis, Dmitriy,

I do not think the behavior should be in any way specific to the case when
persistence is enabled. We have quite a bit of other information (such as
binary marshaller metadata and marshaller class mappings) that are also
stored in the work folder and may be required after the cluster restart,
moreover, this has been so for years since ignite 1.0.

If you think this is a blocker, please suggest what behavior we want to see
in ignite 2.4. We cannot calculate IGNITE_HOME when Ignite is run in an
embedded mode (in other words, the Ignite distribution package structure is
not preserved). Printing a warning in this case looks good enough for me.

2018-01-30 23:35 GMT+03:00 Denis Magda <dmagda@apache.org>:

> The temp dir is no way to go for the persistence.
>
> I would enforce setting paths for the storage and WAL files if IGNITE_HOME
> can't be calculated and we end up placing the storage files under the
> temp.dir.
>
> Alex G., can you join this thread? Seems that there is an oversight on our
> side that needs to be fixed in 2.4.
>
> —
> Denis
>
> > On Jan 30, 2018, at 2:52 AM, Stanislav Lukyanov <stanlukyanov@gmail.com>
> wrote:
> >
> > I checked the code handling the IGNITE_HOME and persistent storage paths,
> > and here is what the algorithm looks like.
> >
> > For IGNITE_HOME the following is checked in order; if on any step a
> value is found then we use it.
> > - IgniteHome in IgniteConfiguration
> > - IGNITE_HOME system property
> > - IGNITE_HOME environment variable
> > - Current working directory (user.dir) and all its ancestors (all
> directories are checked to have “bin/” and “config/”)
> > - Class path entry containing ignite-core classes and all its ancestors
> >
> > After that, the working directory will be created at one of the
> following paths
> > - WorkingDirectory in IgniteConfiguration, if set
> > - ${IGNITE_HOME}/work, if IGNITE_HOME could be calculated previously
> > - ${java.io.tmpdir}/work
> >
> > Persistent storage will be stored in the working directory, unless
> StoragePath are specified in the config
> > (same for WAL and WalPath).
> >
> > The issue here is that if we’ve ended up having persistent DB in the
> working directory in the /tmp,
> > then persistence files will be cleared upon restart.
> > Also, IgniteConfiguration::getIgniteHome claims that Igntie fails if
> IGNITE_HOME is not set, but that’s not the case.
> >
> > So, how about actually disallowing to run Ignite when IGNITE_HOME can’t
> be calculated? Using /tmp for working
> > directory seems to be an obscure and potentially harmful scenario.
> > IgniteConfiguration’s documentation can also be adjusted to specify
> actual steps used to find IgniteHome and WorkingDirectory
> > if they aren’t set explicitly.
> > Additionally, I’d suggest not to promote setting IGNITE_HOME system
> property and environment variable
> > (e.g. let’s remove it from readmeio). IgniteConfiguration seems to be
> the most straightforward way to configure Ignite,
> > and system properties should be used as a backup plan when convenient.
> >
> > WDYT?
> >
> > Thanks,
> > Stan
> >
> > From: Denis Magda
> > Sent: 30 января 2018 г. 3:38
> > To: dev@ignite.apache.org
> > Subject: Re: IGNITE_HOME for persistence
> >
> > No we don’t. I’ve never touched IGNITE_HOME variable for any other
> purpose.
> >
> > As it was suggested, the reported should share the project to reproduce
> his scenario.
> >
> > —
> > Denis
> >
> >> On Jan 26, 2018, at 9:05 PM, Dmitriy Setrakyan <d@gridgain.com> wrote:
> >>
> >> Igniters,
> >>
> >> I have just stumbled upon this post on SO:
> >> https://stackoverflow.com/questions/48434929/apache-
> ignite-persistent-storage
> >>
> >> Do we require IGNITE_HOME to be set if the persistence is enabled? If
> yes,
> >> do we check for it on startup?
> >>
> >> D.
> >
> >
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message