gaia disk usage analysis

readme

Build Tendermint with the diffs in this gist
Run the script to start gaia, collect du, run compact command:
./compact_n <rpc_port> <home> <every_n_blocks>
for example to collect data and compact every 1000 blocks:
./compact_n 26657 ~/.gaia 1000

Plot and check the results:
Rscript plot.R ; open Rplots.pdf

compact_n.sh

function record_du {
    height=$1
    data_file=$2
    csv_file=$3
    du -k $home/data/$data_file
    du=`du -k $home/data/$data_file | head -n1 | awk '{print $1;}'`
    echo -e "$height,$du" >> $home/$csv_file
}

function compact {
    height=$1
    #start the compaction routine
    echo -e "\nCompact at $height"
    pkill gaiad
    sleep 1
    tendermint experimental-compact-goleveldb --home $home > /dev/null 2>&1

    gaiad --home $home start --log_level error > $logfile 2>&1 &
    sleep 2

    #take disk usage after gaia start otherwise the space is not really freed
    du -k $home/data/application.db
    du -k $home/data/blockstore.db
    du -k $home/data/state.db
    du -k $home/data/tx_index.db
    echo "----------------"
}

# ./compact_n <port> <home> <every_n_blocks>
port=$1
home=$2
everyn=$3

#cleanup
pkill gaiad
tendermint experimental-compact-goleveldb --home $home > /dev/null 2>&1

#start gaia
logfile=$home"/gaia.log"
gaiad --home $home start --log_level error > $logfile 2>&1 &
sleep 1

#get current height
height="$(curl -s localhost:$port/status | jq -r .result.sync_info.latest_block_height)"

#get the height's <n> quotient
base=$(($height/$everyn))

#du file header
echo -e "height,usage" > $home/application_du.csv
echo -e "height,usage" > $home/blockstore_du.csv
echo -e "height,usage" > $home/state_du.csv
echo -e "height,usage" > $home/tx_index_du.csv

for i in {1..100}
do
    #determine the compaction height
    compheight=$((($base+$i)*$everyn))

    #wait for the compaction height to be reached
    #collect du at each height
    echo "height is $height"
    while [ $height -lt $compheight ]
    do
        sleep 1
        height="$(curl -s localhost:$port/status | jq -r .result.sync_info.latest_block_height)"

        echo -e "\nDu at $height"
        record_du $height "application.db" "application_du.csv"
        record_du $height "blockstore.db" "blockstore_du.csv"
        record_du $height "state.db" "state_du.csv"
        record_du $height "tx_index.db" "tx_index_du.csv"
        echo -e "----------------"
    done

    compact $height
done

plot.R

app=read.csv("application_du.csv")
bl=read.csv("blockstore_du.csv")
state=read.csv("state_du.csv")
tx_index=read.csv("tx_index_du.csv")
plot(app,type ="l", main="application.db")
plot(bl,type ="l", main="blockstore.db")
plot(state, type="l", main="state.db")
plot(tx_index, type="l", main="tx_index.db")

tendermint diffs

diff --git a/cmd/tendermint/commands/compact.go b/cmd/tendermint/commands/compact.go
index 4813cf1b3..634342583 100644
--- a/cmd/tendermint/commands/compact.go
+++ b/cmd/tendermint/commands/compact.go
@@ -17,8 +17,8 @@ var CompactGoLevelDBCmd = &cobra.Command{
        Use:   "experimental-compact-goleveldb",
        Short: "force compacts the tendermint storage engine (only GoLevelDB supported)",
        Long: `
-This is a temporary utility command that performs a force compaction on the state
-and blockstores to reduce disk space for a pruning node. This should only be run
+This is a temporary utility command that performs a force compaction on the state
+and blockstores to reduce disk space for a pruning node. This should only be run
 once the node has stopped. This command will likely be omitted in the future after
 the planned refactor to the storage engine.

@@ -35,7 +35,7 @@ Currently, only GoLevelDB is supported.
 }

 func compactGoLevelDBs(rootDir string, logger log.Logger) {
-       dbNames := []string{"state", "blockstore"}
+       dbNames := []string{"state", "blockstore", "tx_index", "application"}
        o := &opt.Options{
                DisableSeeksCompaction: true,
        }