I have been using KNIME 2.1.2 with some complex processes in meta-nodes inside loops. The loops iterate over input tables. With the newer versions 2.2.2 and higher the meta-nodes inside the loops are “cached” - that is they are not re-executed for each iteration and return the result from the first loop for each subsequent loops.
Is this a bug, or is there a new setting that causes the results to be cached in the newer versions?
We discovered an ugly bug a few weeks ago, which will be fixed in 2.3.1 (coming out in a few weeks). If the nodes inside the metanode inside the loop are ordered (order of insertion!) in a different way than the execution order then sometimes (under strange side conditions) not all nodes get reset which results in this odd "caching" behaviour since the final nodes of the metanode are not re-executed at all.
Could you try to re-create the part inside the metanode and make sure that you add nodes from left to right?
Sorry for that. Most workflows are created mostly left-to-right and this really only affects meta nodes inside loops so we never came across this in any of our test cases.
I removed and replaced nodes in the flow, would that affect this ordering? That is, I deleted nodes from the middle of the flow, then put in new ones, or in some cases inserted nodes to re-order columns in the workflow after the initial building of the meta-node. Would that cause the problem?
The editing scenario seems like it would happen a lot for serious developments projects.