geth fast sync restarts at end of blockchain

[adsmaniotto]

System information

I'm running Mist version 0.7.4 on Windows 10.

Expected behaviour

I have been trying for over a week to sync the blockchain using fast sync but have been running into an issue where the sync appears to restart itself when there's only a few thousand blocks left to go. It appears to roll back a couple thousands headers around Block 1606908, and then starts the sync again with a new peer (albiet at a much slower speed).

The issue seems related to Issue #2469. Is there any available workaround? The fast sync command below took 8 hours to get near the end before the restart.

Backtrace

geth --fast --cache=1024 --jitvm

[karalabe]

Try latest develop, or if your adventurous try #2630. I can fast sync in 25-30 mins flat (albeit on a 200mbit link). The PR I mentioned could especially help on poorer connections.

[karalabe]

This sync issue should be solved by our 1.4.6 release. Please check with that and report if it's still broken for you.

[ghost]

I hope it's appropriate for me to re-open this.

I'm on 1.4.6 on Ubuntu 16.04

When I delete ~/.ethereum/chaindata/*

And re-start geth:

geth --fast console

Days pass...

Everything fine...

Then I Ctrl+D in geth.

A few moments later I restart:

geth --fast console

fast sync disabled! I have to delete everything in chaindata and restart...

I0621 23:19:28.560522 core/blockchain.go:208] Fast block: #1747043 [bd2635c4…] TD=29636013546513671717
I0621 23:19:28.578204 eth/handler.go:110] blockchain not empty, fast sync disabled
I0621 23:19:28.581229 p2p/server.go:313] Starting Server

[fjl]

That's intended. Fast sync runs only once for security reasons.

[fjl]

This issue was specifically about a bug where fast sync would restart a normal sync from block zero just after completing.

[ghost]

Apologies for confusing the issue. Thanks for explaining about the one-time fast sync'ing for security reasons. Very helpful information.

[ModGrip]

How do I know when fast sync has completed? Appears to still be running at higher than the current block number. "Processed 3660436" currently and rising.

[Arthur111]

That's intended. Fast sync runs only once for security reasons.

fast sync must be disabled when the first time its sync.
Until its not sync the first time --fast must stay enabled.
Like this you can restart as you want or after a crash (bug) until its sync for the first time.