core/state/snapshot: avoid thrashing during block import

[holiman]

The code in this PR is written by @karalabe (I just moved a chunk of code a bit north)
When we're generating the snapshot, we're also processing blocks. In case we're importing multiple blocks (catching up), this adds a penalty, since there are background processes which are reading a lot from disk, all the while we're trying to catch up.

This PR instead makes it so that if the previous generation-effort ended within 450ms, then it won't start anew.

[holiman]

TODO: I haven't tested it on an actual sync, will try that in the near future.

[karalabe]

This is not enough though. That's why I didn't open it as a PR. Currently generation runs during block processing too, not just in between, that's the primary issue. Until that's fixed, generation - especially with cold caches on startup - will cause block processing to take seconds, which makes this check a noop.

[holiman]

A naive but straight-forward approach:

diff --git a/core/blockchain.go b/core/blockchain.go
index 00f90415c1..bf632ae32e 100644
--- a/core/blockchain.go
+++ b/core/blockchain.go
@@ -1772,6 +1772,8 @@ func (bc *BlockChain) insertChain(chain types.Blocks, verifySeals bool) (int, er
 			activeState.StopPrefetcher()
 		}
 	}()
+	// Pause the snapshot generator, if it's working
+	bc.snaps.PauseGenerator()
 
 	for ; block != nil && err == nil || err == ErrKnownBlock; block, err = it.next() {
 		// If the chain is terminating, stop processing blocks
diff --git a/core/state/snapshot/snapshot.go b/core/state/snapshot/snapshot.go
index 8889fa9fec..25eab2c70a 100644
--- a/core/state/snapshot/snapshot.go
+++ b/core/state/snapshot/snapshot.go
@@ -233,6 +233,17 @@ func (t *Tree) waitBuild() {
 		<-done
 	}
 }
+func (t *Tree) PauseGenerator() {
+	t.lock.Lock()
+	defer t.lock.Unlock()
+	layer := t.disklayer()
+	// If the base layer is generating, abort it
+	if layer.genAbort != nil {
+		abort := make(chan *generatorStats)
+		layer.genAbort <- abort
+		<-abort
+	}
+}
 
 // Disable interrupts any pending snapshot generator, deletes all the snapshot
 // layers in memory and marks snapshots disabled globally. In order to resume

Testing it a bit now, not sure if I'm missing something

[Cr4shOv3rrid3]

can someone plz fix this. i am having issues because of that with a virtual private server setup that is not allowing me to make a new snapshot. seem like it also affects the bitcoin course price already. those pull requests in both networks always synch into latest course price. at least we get a roof that finally makes sense now for bitcoin tech for price speculations (no.of latest pull request). guess this makes the price of 1 ETH (this one included and herby hardcoded into this comment). round about a net income on a monthly basis, from an average user for this tech (what would also make the most sense and would put bitcoin into net income for lowest connected people per year as i think the savings aspect bitcoin is the better choice cause ethereum wasnt build by savings people. and savings commonly are APY based.)

besides of this i would ACK the idea for the fix that holiman already proposed.
switching vom a into b chain as well sounds reasonable (bitcoin consensus coded).

change my comment or delete it according to your needs if you want. i am not a core developer but i am working with the code of both networks a lot. so i just wanna help this tech to run more smoothly and for what it originally came to life.