setStorageAddresses support to optionally override storage addresses

[wey-gu]

This enables vesoft-inc/nebula-spark-connector#39 , where we could make spark projects(spark-c, exchange, algorithm) running outside namespace/cluster of Nebula Graph Cluster in K8s being work together with Nebula(including the DBaaS).

[codecov-commenter]

Codecov Report

Merging #451 (602a904) into master (a8329e2) will decrease coverage by 0.44%.
The diff coverage is 66.66%.

@@             Coverage Diff              @@
##             master     #451      +/-   ##
============================================
- Coverage     63.75%   63.30%   -0.45%     
+ Complexity      736      729       -7     
============================================
  Files            66       66              
  Lines          3479     3483       +4     
  Branches        507      506       -1     
============================================
- Hits           2218     2205      -13     
- Misses          920      931      +11     
- Partials        341      347       +6     

Impacted Files	Coverage Δ
...om/vesoft/nebula/client/storage/StorageClient.java	73.73% <66.66%> (-1.53%)	⬇️
...java/com/vesoft/nebula/client/meta/MetaClient.java	52.88% <0.00%> (-4.81%)	⬇️
.../client/storage/scan/ScanVertexResultIterator.java	62.63% <0.00%> (-2.20%)	⬇️
...t/nebula/client/storage/scan/ScanVertexResult.java	64.61% <0.00%> (-1.54%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a8329e2...602a904. Read the comment docs.

[HarrisChu]

I don't think so.
If this way, how to verify if the host is valid?

the same question:

when deploy HDFS in k8s, how to use spark outside k8s?

[HarrisChu]

another question, when scan data from storage, we must know which host the leader in

if user provides the hosts, we cannot map the partitions in meta.

[wey-gu]

Thanks @HarrisChu !

I didn't realized that we actually need the mapping of meta-storage in many spark cases, was just consider it's feasible as we have similar interfaces of storage client in nebula-python.

In this case, to make Nebula in k8s work with Spark, when the spark is not running inside of the NebulaCluster Namespace, the only way is to hack on:

a.

• expose meta/storage endpoint and remaining same port
• use dns to enable same domain name to be resolved to it

b.
use reverse proxy to mock the endpoints with same domain name and port(this decouple the requirement of same port).

I'll try b. first as a poc/reference then.
PoC here: https://gist.github.com/wey-gu/699b9a2ef5dff5f0fb5f288d692ddfd5

Thanks!

[HarrisChu]

@MegaByte875
CC, may be a feature in operator

[wey-gu]

Or, we should introduce withStorageAddress in map instead of list(current implementation), where the specified storage address and its actually address in metad's mapping relationship should be provided.

[Nicole00]

close for more discuss for exposing the storaged address.