A quick question, I have only enabled Pinot Server...
# generalm
Mannoj
04/18/2025, 8:07 AMA quick question, I have only enabled Pinot Servers without HDFS. And data has been populating on them. Now when I enable HDFS, how the data sends to HDFS?
1. Will it automatically send data into HDFS
2. Or Pinot-server and pinot-controllers restart required?
3. Or rebalance or reload table will do the magic of sending data into hdfs?
4. NOPE : Nothing will do, after enabling hdfs. you need to drop table and recreate on pinot so it will start sending data to pinot and hdfs.
x
Xiang Fu
04/18/2025, 10:15 AM
do you already use hdfs for deep store or it’s using controller disk?
Xiang Fu
04/18/2025, 10:15 AM
if you are already using hdfs for controller deep store, then you don’t need to do anything
Xiang Fu
04/18/2025, 10:16 AM
otherwise, all old data will be stay where it is, but new data pushed will be on hdfs
Xiang Fu
04/18/2025, 10:17 AM
OR you need to move data to deep store and modify the segment metadata yourself
m
Mannoj
04/18/2025, 1:39 PMGot it thanks
Mannoj
04/18/2025, 1:44 PMTo correct my understanding
• if controller deep store disks are in hdfs, then it will push data to server deepstorage also for whatever data pinot server has. Since it has metastore information it can send it to hdfs deep store newly configured on pinot servers. Right?
x
Xiang Fu
04/18/2025, 1:50 PM
pinot uses deep store for data backup.
Xiang Fu
04/18/2025, 1:50 PM
both controller and server should use same deep store
Xiang Fu
04/18/2025, 1:51 PM
before: realtime segments pushed from pinot server -> controller -> hdfs
new: segments are pushed from server -> hdfs
m
Mannoj
04/18/2025, 1:58 PMAh so if pinot server was using controller deepstore before then it is possible, well my query is i never enabled hdfs on both controller nor pinotserver , first time i enabled on controller and server and did rolling restart on them. In this case how i can backfill existing pinot data into hdfs?
x
Xiang Fu
04/18/2025, 2:24 PM
you need to manually copy those segments from controller deep store to hdfs
m
Mannoj
04/18/2025, 4:45 PMController deep store just has metastore segments? Not actual pinot data. Am I right?
x
Xiang Fu
04/18/2025, 4:57 PM
deep store has compressed data, used as backup;
metastore is in zookeeper
m
Mannoj
04/21/2025, 7:24 AMGood Morning, was out on weekend,
your response:
you need to manually copy those segments from controller deep store to hdfsx
Xiang Fu
04/21/2025, 7:30 AM
hmm, then what’s the download url for the segment in your segment metadata? could you please check
m
Mannoj
04/21/2025, 8:20 AMlocal /data/controller
x
Xiang Fu
04/21/2025, 9:02 AM
what’s the sample url
m
Mannoj
04/21/2025, 9:09 AMSomething like this.
# Pinot Controller VIP Port
controller.vip.port=9000
# Location to store Pinot Segments pushed from clients
controller.data.dir=/data/pinot
x
Xiang Fu
04/21/2025, 9:45 AM
oh, I mean go to UI, check cluster and find any table , find a segment, and check segment metadata, there is a field segment.download.url
m
Mannoj
04/21/2025, 10:48 AMright now since I have hdfs enabled so it shows : "segment.download.url": "hdfs://pinotpoc/user/pinot/airlineStats/airlineStats_16082_16082_0"
Earlier this must be null , since I never had a hdfs layer.
x
Xiang Fu
04/21/2025, 12:47 PM
ok, can you check older segments metadata?
Xiang Fu
04/21/2025, 12:47 PM
new segments are hdfs then you are good to go
Xiang Fu
04/21/2025, 12:50 PM
so the steps are:
1. find old segments download url, -> find where those segments tar file physically located
2. copy them to hdfs
3. update segment metadata download url to the hdfs url
2 Views