forked from D-Net/dnet-hadoop
219 lines
12 KiB
Bash
219 lines
12 KiB
Bash
export PYTHON_EGG_CACHE=/home/$(whoami)/.python-eggs
|
|
export link_folder=/tmp/impala-shell-python-egg-cache-$(whoami)
|
|
if ! [ -L $link_folder ]
|
|
then
|
|
rm -Rf "$link_folder"
|
|
ln -sfn ${PYTHON_EGG_CACHE}${link_folder} ${link_folder}
|
|
fi
|
|
|
|
export HADOOP_USER_NAME=$2
|
|
|
|
# Set the active HDFS node of OCEAN and IMPALA cluster.
|
|
OCEAN_HDFS_NODE='hdfs://nameservice1'
|
|
echo -e "\nOCEAN HDFS virtual-name which resolves automatically to the active-node: ${OCEAN_HDFS_NODE}"
|
|
|
|
IMPALA_HDFS_NODE=''
|
|
COUNTER=0
|
|
while [ $COUNTER -lt 3 ]; do
|
|
if hdfs dfs -test -e hdfs://impala-cluster-mn1.openaire.eu/tmp >/dev/null 2>&1; then
|
|
IMPALA_HDFS_NODE='hdfs://impala-cluster-mn1.openaire.eu:8020'
|
|
break
|
|
elif hdfs dfs -test -e hdfs://impala-cluster-mn2.openaire.eu/tmp >/dev/null 2>&1; then
|
|
IMPALA_HDFS_NODE='hdfs://impala-cluster-mn2.openaire.eu:8020'
|
|
break
|
|
else
|
|
IMPALA_HDFS_NODE=''
|
|
sleep 1
|
|
fi
|
|
((COUNTER++))
|
|
done
|
|
if [ -z "$IMPALA_HDFS_NODE" ]; then
|
|
echo -e "\n\nERROR: PROBLEM WHEN SETTING THE HDFS-NODE FOR IMPALA CLUSTER! | AFTER ${COUNTER} RETRIES.\n\n"
|
|
exit 1
|
|
fi
|
|
echo -e "Active IMPALA HDFS Node: ${IMPALA_HDFS_NODE} , after ${COUNTER} retries.\n\n"
|
|
|
|
IMPALA_HOSTNAME='impala-cluster-dn1.openaire.eu'
|
|
IMPALA_CONFIG_FILE='/etc/impala_cluster/hdfs-site.xml'
|
|
|
|
IMPALA_HDFS_DB_BASE_PATH="${IMPALA_HDFS_NODE}/user/hive/warehouse"
|
|
|
|
|
|
# Set sed arguments.
|
|
LOCATION_HDFS_NODE_SED_ARG="s|${OCEAN_HDFS_NODE}|${IMPALA_HDFS_NODE}|g" # This requires to be used with "sed -e" in order to have the "|" delimiter (as the "/" conflicts with the URIs)
|
|
|
|
# Set the SED command arguments for column-names with reserved words:
|
|
DATE_SED_ARG_1='s/[[:space:]]\date[[:space:]]/\`date\`/g'
|
|
DATE_SED_ARG_2='s/\.date,/\.\`date\`,/g' # the "date" may be part of a larger field name like "datestamp" or "date_aggregated", so we need to be careful with what we are replacing.
|
|
DATE_SED_ARG_3='s/\.date[[:space:]]/\.\`date\` /g'
|
|
|
|
HASH_SED_ARG_1='s/[[:space:]]\hash[[:space:]]/\`hash\`/g'
|
|
HASH_SED_ARG_2='s/\.hash,/\.\`hash\`,/g'
|
|
HASH_SED_ARG_3='s/\.hash[[:space:]]/\.\`hash\` /g'
|
|
|
|
LOCATION_SED_ARG_1='s/[[:space:]]\location[[:space:]]/\`location\`/g'
|
|
LOCATION_SED_ARG_2='s/\.location,/\.\`location\`,/g'
|
|
LOCATION_SED_ARG_3='s/\.location[[:space:]]/\.\`location\` /g'
|
|
|
|
|
|
function copydb() {
|
|
db=$1
|
|
echo -e "\nStart processing db: '${db}'..\n"
|
|
|
|
# Delete the old DB from Impala cluster (if exists).
|
|
impala-shell --user ${HADOOP_USER_NAME} -i ${IMPALA_HOSTNAME} -q "drop database if exists ${db} cascade" |& tee error.log # impala-shell prints all logs in stderr, so wee need to capture them and put them in a file, in order to perform "grep" on them later
|
|
log_errors=`cat error.log | grep -E "WARN|ERROR|FAILED"`
|
|
if [ -n "$log_errors" ]; then
|
|
echo -e "\n\nERROR: THERE WAS A PROBLEM WHEN DROPPING THE OLD DATABASE! EXITING...\n\n"
|
|
rm -f error.log
|
|
exit 2
|
|
fi
|
|
|
|
echo -e "\n\nCopying files of '${db}', from Ocean to Impala cluster..\n"
|
|
# Using max-bandwidth of: 70 * 150 Mb/s = 10.5 Gb/s
|
|
# Using max memory of: 70 * 6144 = 430 Gb
|
|
# Using 1MB as a buffer-size.
|
|
# The " -Ddistcp.dynamic.recordsPerChunk=N" arg is not available in our version of hadoop
|
|
# The "ug" args cannot be used as we get a "User does not belong to hive" error.
|
|
# The "p" argument cannot be used, as it blocks the files from being used, giving a "sticky bit"-error, even after applying chmod and chown onm the files.
|
|
hadoop distcp -Dmapreduce.map.memory.mb=6144 -m 70 -bandwidth 150 \
|
|
-numListstatusThreads 40 \
|
|
-copybuffersize 1048576 \
|
|
-strategy dynamic \
|
|
-blocksperchunk 8 \
|
|
-pb \
|
|
${OCEAN_HDFS_NODE}/user/hive/warehouse/${db}.db ${IMPALA_HDFS_DB_BASE_PATH}
|
|
|
|
# Check the exit status of the "hadoop distcp" command.
|
|
if [ $? -eq 0 ]; then
|
|
echo -e "\nSuccessfully copied the files of '${db}'.\n"
|
|
else
|
|
echo -e "\n\nERROR: FAILED TO TRANSFER THE FILES OF '${db}', WITH 'hadoop distcp'. GOT EXIT STATUS: $?\n\n"
|
|
rm -f error.log
|
|
exit 3
|
|
fi
|
|
|
|
# In case we ever use this script for a writable DB (using inserts/updates), we should perform the following costly operation as well..
|
|
#hdfs dfs -conf ${IMPALA_CONFIG_FILE} -chmod -R 777 ${TEMP_SUBDIR_FULLPATH}/${db}.db
|
|
|
|
echo -e "\nCreating schema for db: '${db}'\n"
|
|
|
|
# create the new database (with the same name)
|
|
impala-shell --user ${HADOOP_USER_NAME} -i ${IMPALA_HOSTNAME} -q "create database ${db}"
|
|
|
|
# Because "Hive" and "Impala" do not have compatible schemas, we cannot use the "show create table <name>" output from hive to create the exact same table in impala.
|
|
# So, we have to find at least one parquet file (check if it's there) from the table in the ocean cluster for impala to use it to extract the table-schema itself from that file.
|
|
|
|
all_create_view_statements=()
|
|
num_tables=0
|
|
|
|
entities_on_ocean=`hive -e "show tables in ${db};" | sed 's/WARN:.*//g'` # Get the tables and views without any potential the "WARN" logs.
|
|
for i in ${entities_on_ocean[@]}; do # Use un-quoted values, as the elemetns are single-words.
|
|
# Check if this is a view by showing the create-statement where it should print "create view" for a view, not the "create table". Unfortunately, there is no "show views" command.
|
|
create_entity_statement=`hive -e "show create table ${db}.${i};"` # It needs to happen in two stages, otherwise the "grep" is not able to match multi-line statement.
|
|
|
|
create_view_statement_test=`echo -e "$create_entity_statement" | grep 'CREATE VIEW'`
|
|
if [ -n "$create_view_statement_test" ]; then
|
|
echo -e "\n'${i}' is a view, so we will save its 'create view' statement and execute it on Impala, after all tables have been created.\n"
|
|
create_view_statement=`echo -e "$create_entity_statement" | sed 's/WARN:.*//g' | sed 's/\`//g' \
|
|
| sed 's/"$/;/' | sed 's/^"//' | sed 's/\\"\\"/\"/g' | sed -e "${LOCATION_HDFS_NODE_SED_ARG}" | sed "${DATE_SED_ARG_1}" | sed "${HASH_SED_ARG_1}" | sed "${LOCATION_SED_ARG_1}" \
|
|
| sed "${DATE_SED_ARG_2}" | sed "${HASH_SED_ARG_2}" | sed "${LOCATION_SED_ARG_2}" \
|
|
| sed "${DATE_SED_ARG_3}" | sed "${HASH_SED_ARG_3}" | sed "${LOCATION_SED_ARG_3}"`
|
|
all_create_view_statements+=("$create_view_statement")
|
|
else
|
|
echo -e "\n'${i}' is a table, so we will check for its parquet files and create the table on Impala cluster.\n"
|
|
((num_tables++))
|
|
CURRENT_PRQ_FILE=`hdfs dfs -conf ${IMPALA_CONFIG_FILE} -ls -C "${IMPALA_HDFS_DB_BASE_PATH}/${db}.db/${i}/" | grep -v 'Found' | grep -v '_impala_insert_staging' | head -1`
|
|
if [ -z "$CURRENT_PRQ_FILE" ]; then # If there is not parquet-file inside.
|
|
echo -e "\nERROR: THE TABLE \"${i}\" HAD NO FILES TO GET THE SCHEMA FROM! IT'S EMPTY!\n\n"
|
|
exit 4 # Comment out when testing a DB which has such a table, just for performing this exact test-check.
|
|
else
|
|
impala-shell --user ${HADOOP_USER_NAME} -i ${IMPALA_HOSTNAME} -q "create table ${db}.${i} like parquet '${CURRENT_PRQ_FILE}' stored as parquet;" |& tee error.log
|
|
log_errors=`cat error.log | grep -E "WARN|ERROR|FAILED"`
|
|
if [ -n "$log_errors" ]; then
|
|
echo -e "\n\nERROR: THERE WAS A PROBLEM WHEN CREATING TABLE '${i}'!\n\n"
|
|
fi
|
|
fi
|
|
fi
|
|
done
|
|
|
|
previous_num_of_views_to_retry=${#all_create_view_statements[@]}
|
|
if [[ $num_tables -gt 0 ]]; then
|
|
echo -e "\nAll ${num_tables} tables have been created, for db '${db}', going to create the ${previous_num_of_views_to_retry} views..\n"
|
|
else
|
|
echo -e "\nDB '${db}' does not have any tables, moving on to create the ${previous_num_of_views_to_retry} views..\n"
|
|
fi
|
|
|
|
if [[ $previous_num_of_views_to_retry -gt 0 ]]; then
|
|
echo -e "\nAll_create_view_statements (${previous_num_of_views_to_retry}):\n\n${all_create_view_statements[@]}\n" # DEBUG
|
|
else
|
|
echo -e "\nDB '${db}' does not contain any views.\n"
|
|
fi
|
|
|
|
level_counter=0
|
|
while [[ $previous_num_of_views_to_retry -gt 0 ]]; do
|
|
((level_counter++))
|
|
# The only accepted reason for a view to not be created, is if it depends on another view, which has not been created yet.
|
|
# In this case, we should retry creating this particular view again.
|
|
new_num_of_views_to_retry=0
|
|
|
|
for create_view_statement in "${all_create_view_statements[@]}"; do # Here we use double quotes, as the elements are phrases, instead of single-words.
|
|
impala-shell --user ${HADOOP_USER_NAME} -i ${IMPALA_HOSTNAME} -q "${create_view_statement}" |& tee error.log # impala-shell prints all logs in stderr, so wee need to capture them and put them in a file, in order to perform "grep" on them later
|
|
specific_errors=`cat error.log | grep -E "FAILED: ParseException line 1:13 missing TABLE at 'view'|ERROR: AnalysisException: Could not resolve table reference:"`
|
|
if [ -n "$specific_errors" ]; then
|
|
echo -e "\nspecific_errors: ${specific_errors}\n"
|
|
echo -e "\nView '$(cat error.log | grep -Eo "Query: CREATE VIEW ([^\s]+)" | sed 's/Query: CREATE VIEW //g')' failed to be created, possibly because it depends on another view.\n"
|
|
((new_num_of_views_to_retry++)) # Increment it here, instead of acquiring the array's size in the end, as that doesn't work for some reason.
|
|
else
|
|
all_create_view_statements=("${all_create_view_statements[@]/$create_view_statement}") # Remove the current successful statement from the list.
|
|
sleep 1 # Wait a bit for Impala to register that the view was created, before possibly referencing it by another view.
|
|
fi
|
|
done
|
|
|
|
all_create_view_statements=("$(echo "${all_create_view_statements[@]}" | grep -v '^[\s]*$')") # Re-index the array, filtering-out any empty elements.
|
|
# Although the above command reduces the "active" elements to just the few to-be-retried, it does not manage to make the array return the its true size through the "${#all_create_view_statements[@]}" statement. So we use counters.
|
|
|
|
if [[ $new_num_of_views_to_retry -eq $previous_num_of_views_to_retry ]]; then
|
|
echo -e "\n\nERROR: THE NUMBER OF VIEWS TO RETRY HAS NOT BEEN REDUCED! THE SCRIPT IS LIKELY GOING TO AN INFINITE-LOOP! EXITING..\n\n"
|
|
exit 5
|
|
elif [[ $new_num_of_views_to_retry -gt 0 ]]; then
|
|
echo -e "\nTo be retried \"create_view_statements\" (${new_num_of_views_to_retry}):\n\n${all_create_view_statements[@]}\n"
|
|
else
|
|
echo -e "\nFinished creating views, for db: '${db}', in level-${level_counter}.\n"
|
|
fi
|
|
previous_num_of_views_to_retry=$new_num_of_views_to_retry
|
|
done
|
|
|
|
echo -e "\nComputing stats for tables..\n"
|
|
entities_on_impala=`impala-shell --user ${HADOOP_USER_NAME} -i ${IMPALA_HOSTNAME} --delimited -q "show tables in ${db}"`
|
|
for i in ${entities_on_impala[@]}; do # Use un-quoted values, as the elemetns are single-words.
|
|
# Taking the create table statement from the Ocean cluster, just to check if its a view, as the output is easier than using impala-shell from Impala cluster.
|
|
create_view_statement=`hive -e "show create table ${db}.${i};" | grep "CREATE VIEW"` # This grep works here, as we do not want to match multiple-lines.
|
|
if [ -z "$create_view_statement" ]; then # If it's a table, then go load the data to it.
|
|
# Invalidate metadata of this DB's tables, in order for Impala to be aware of all parquet files put inside the tables' directories, previously, by "hadoop distcp".
|
|
impala-shell --user ${HADOOP_USER_NAME} -i ${IMPALA_HOSTNAME} -q "INVALIDATE METADATA ${db}.${i}"
|
|
sleep 1
|
|
impala-shell --user ${HADOOP_USER_NAME} -i ${IMPALA_HOSTNAME} -q "compute stats ${db}.${i}";
|
|
fi
|
|
done
|
|
|
|
# Check if the entities in both clusters are the same, down to the exact names, not just the counts. (they are sorted in the same way both in hive and impala)
|
|
if [ "${entities_on_impala[@]}" == "${entities_on_ocean[@]}" ]; then
|
|
echo -e "\nAll entities have been copied to Impala cluster.\n"
|
|
else
|
|
echo -e "\n\nERROR: 1 OR MORE ENTITIES OF DB '${db}' FAILED TO BE COPIED TO IMPALA CLUSTER!\n\n"
|
|
rm -f error.log
|
|
exit 6
|
|
fi
|
|
|
|
rm -f error.log
|
|
echo -e "\n\nFinished processing db: ${db}\n\n"
|
|
}
|
|
|
|
|
|
MONITOR_DB=$1
|
|
|
|
copydb $MONITOR_DB'_institutions'
|
|
copydb $MONITOR_DB
|
|
|