Taras Ledkov created IGNITE-13154:
-------------------------------------
Summary: Introduce aility to manage manage binary types by the
users
Key: IGNITE-13154
URL: https://issues.apache.org/jira/browse/IGNITE-13154
Project: Ignite
Issue Type: Improvement
Components: binary
Reporter: Taras Ledkov
Assignee: Taras Ledkov
Fix For: 2.9
We need a way to change schema (including unsupported changes, such as column
type change) without cluster restart. This is for the case when all data
associated with the binary type has been removed, so removing the old schema is
safe.
Now users must stop the cluster and remove the folder with binary metadata
manually.
The proposed way is to introduce internal API to manage binary types and public
command line interface (via control.sh). That way one can remove a cache from
the cluster, then unregister corresponding binary types, then launch a new
version of an application that would register the new schema and reload the
data.
*The current implementation has restrictions:*
- all cluster nodes must support remove type feature.
- the cluster must not contains data with type to remove.
- operation of the update type must not be launched on the cluster for the type
to remove (operation examples: put into cache, BinaryObjectBuilder#build).
- client nodes process metadata operation asynchronously so type may be removed
at the client after any delay after the remove type operation is completed.
- if the node that contains the old version of the type joins to the cluster
where type was removed the type is propagated to cluster metadata (because
metadata tombstone not supported).
- if the node that contains the old version of the type cannot join to the
cluster where type was removed and then updated to the new version (because
metadata versioned tombstone not supported).
So, user scenarios looks like:
# Be sure that all server nodes supports remove type feature.
# Remove caches contains the data with type to remove.
# Stops the client node with older version.
# Stops all operation with type to remove (don't create binary objects, don't
run compute jobs with type to remove).
# Remove the type on the stable topology (and production destination topolog).
# Waits any delay (depends on the cluster size and clients count)
# Produce operations with new version of the type.
*Proposed command line interface*
New commands (all commands are _experimental_ ):
- {{--meta list}} prints info about all available binary types:
{{typeId=<ID>, typeName=<name>, fields=<fields_count>, schemas=<schemas_count>,
isEnum=<bool>}}
- {{\-\-meta details (\-\- typeId <ID>| \-\-typeName <name>)}} prints detailed
info info about specified type. The type may be specified by type name or type
ID.
output example:
{code}
typeId=0x1FBFBC0C (532659212)
typeName=TypeName1
Fields:
name=fld3, type=long[], fieldId=0x2FFF95 (3145621)
name=fld2, type=double, fieldId=0x2FFF94 (3145620)
name=fld1, type=Date, fieldId=0x2FFF93 (3145619)
name=fld0, type=int, fieldId=0x2FFF92 (3145618)
Schemas:
schemaId=0x6C5CC179 (1818018169), fields=[fld0]
schemaId=0x70E46431 (1894016049), fields=[fld0, fld1, fld2, fld3]
{code}
- {{\-\-meta remove (\-\- typeId <ID>| \-\-typeName <name>) [\-\-out
<file_name>]}} removes metadata for specified type form cluster and saves the
removed metadata to the specified file. If the file name isn't specified the
output file name is: {{<typeId>.bin}}
The command requires confirmation.
*N.B.*: The all session of thin clients (ODBC, JDBC, thin client) are closed
(to cleanup local cache of the binary metadata).
- {{\-\-meta update \-\-in <file_name>]}} update cluster metadata from
specified file (file name is required)
The command requires confirmation.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
