Overview of the Issue
This issue was found while working on: #14424
The workflow update command allows you to update some configuration options for an existing workflow.
For most VReplication workflows there's a single stream per target tablet. In the cases of e.g. shard merges or materializations, however, you can have N streams (1 per source shard).
The workflow update command only updated one stream on each target shard.
Because the workflow start|stop command uses the same UpdateVReplicationWorkflow tablet RPC that the workflow update command does — only updating the state — it is also impacted by this.
In this scenario, where there's more than one stream on the target tablet, the update and start/stop commands simply DO NOTHING — which is reflected in the output as having changed none of the shards involved in the workflow rather than with returning an error.
Reproduction Steps
git checkout main && make build
cd examples/local
./101_initial_cluster.sh; mysql < ../common/insert_commerce_data.sql; ./201_customer_tablets.sh; ./202_move_tables.sh; ./203_switch_reads.sh; ./204_switch_writes.sh; ./205_clean_commerce.sh; ./301_customer_sharded.sh; ./302_new_shards.sh; ./303_reshard.sh; ./304_switch_reads.sh; ./305_switch_writes.sh
# For stop/start
vtctldclient workflow --keyspace customer show --workflow cust2cust_reverse --compact --include-logs=false | grep state
vtctldclient workflow --keyspace customer stop --workflow cust2cust_reverse
vtctldclient workflow --keyspace customer show --workflow cust2cust_reverse --compact --include-logs=false | grep state
# For update
command mysql -u root --socket="${VTDATAROOT}/vt_0000000200/mysql.sock" -e "select cell from _vt.vreplication"
vtctldclient workflow --keyspace customer update --workflow cust2cust_reverse --cells cell1 --cells cell2
command mysql -u root --socket="${VTDATAROOT}/vt_0000000200/mysql.sock" -e "select cell from _vt.vreplication"
Binary Version
vtgate version Version: 19.0.0-SNAPSHOT (Git revision 3037acd816ba6e8cb299e00f478a8555dc0a1ab8 branch 'main') built on Thu Nov 2 19:17:24 EDT 2023 by matt@pslord.local using go1.21.3 darwin/arm64
Operating System and Environment details
Log Fragments
No response
Overview of the Issue
This issue was found while working on: #14424
The
workflow updatecommand allows you to update some configuration options for an existing workflow.For most VReplication workflows there's a single stream per target tablet. In the cases of e.g. shard merges or materializations, however, you can have N streams (1 per source shard).
The workflow update command only updated one stream on each target shard.
Because the
workflow start|stopcommand uses the sameUpdateVReplicationWorkflowtablet RPC that theworkflow updatecommand does — only updating the state — it is also impacted by this.In this scenario, where there's more than one stream on the target tablet, the update and start/stop commands simply DO NOTHING — which is reflected in the output as having changed none of the shards involved in the workflow rather than with returning an error.
Reproduction Steps
Binary Version
vtgate version Version: 19.0.0-SNAPSHOT (Git revision 3037acd816ba6e8cb299e00f478a8555dc0a1ab8 branch 'main') built on Thu Nov 2 19:17:24 EDT 2023 by matt@pslord.local using go1.21.3 darwin/arm64Operating System and Environment details
Log Fragments
No response