Skip to content

Latest commit

 

History

History
2142 lines (1557 loc) · 68.7 KB

README.md

File metadata and controls

2142 lines (1557 loc) · 68.7 KB
Raw

s3utils

S3 Connector and Zenko Utilities

Run the Docker container as (replace <tag> with the tag of s3utils image to be used)

docker run --net=host -e 'ACCESS_KEY=accessKey' -e 'SECRET_KEY=secretKey' -e 'ENDPOINT=http://127.0.0.1:8000' -e 'REPLICATION_GROUP_ID=RG001'
ghcr.io/scality/s3utils:<tag> node scriptName bucket1[,bucket2...]

Trigger CRR on existing objects

  1. Enable versioning and set up replication on the bucket.
  2. Run script as
node crrExistingObjects.js testbucket1,testbucket2

Mandatory environment variables,

  • REPLICATION_GROUP_ID

Optional environment variables,

  • DEBUG: set to 1 to output debug level information.

Extra environment variables

Additionally, the following extra environment variables can be passed to the script to modify its behavior:

TARGET_REPLICATION_STATUS

Comma-separated list of replication statuses to target for CRR requeueing. The recognized statuses are:

  • NEW: No replication status is attached to the object. This is the state of objects written without any CRR policy attached to the bucket that would have triggered CRR on them.

  • PENDING: The object replication status is PENDING.

  • COMPLETED: The object replication status is COMPLETED.

  • FAILED: The object replication status is FAILED.

  • REPLICA: The object replication status is REPLICA (objects that were put to a target site via CRR have this status).

The default script behavior is to affect objects that have no replication status attached (so equivalent to TARGET_REPLICATION_STATUS=NEW).

Examples:

TARGET_REPLICATION_STATUS=PENDING,COMPLETED

Requeue objects that either have a replication status of PENDING or COMPLETED for CRR, do not requeue the others.

TARGET_REPLICATION_STATUS=NEW,PENDING,COMPLETED,FAILED

Trigger CRR on all original source objects (not replicas) in a bucket.

TARGET_REPLICATION_STATUS=REPLICA

For disaster recovery notably, it may be useful to reprocess REPLICA objects to re-sync a backup bucket to the primary site.

TARGET_PREFIX

Optional prefix of keys to scan

Example:

TARGET_PREFIX=foo/

Only scan keys beginning with "foo/"

STORAGE_TYPE

Comma-separated list of the destination storage location types. This is used only if a replication destination is a public cloud.

The recognized statuses are:

  • aws_s3: The destination storage location type is Amazon S3.

  • azure: The destination storage location type is Microsoft Azure Blob Storage.

  • gcp: The destination storage location type is Google Cloud Storage.

Examples:

STORAGE_TYPE=aws_s3

The destination storage location type is AWS.

STORAGE_TYPE=aws_s3,azure,gcp

The destination storage type is a one-to-many configuration replicating to AWS, Azure, and GCP destination storage locations.

WORKERS

Specify how many parallel workers should run to update object metadata. The default is 10 parallel workers.

Example:

WORKERS=50

MAX_UPDATES, MAX_SCANNED

Specify a maximum number of metadata updates (MAX_UPDATES) or scanned entries (MAX_SCANNED) to execute before stopping the script.

If the script reaches this limit, it outputs a log line containing the KeyMarker and VersionIdMarker to pass to the next invocation (as environment variables KEY_MARKER and VERSION_ID_MARKER) and the updated bucket list without the already completed buckets. At the next invocation of the script, those two environment variables must be set and the updated bucket list passed on the command line to resume where the script stopped.

The default is unlimited (will process the complete listing of buckets passed on the command line).

If the script queues too many objects and Backbeat cannot process them quickly enough, Kafka may drop the oldest entries, and the associated objects will stay in the PENDING state permanently without being replicated. When the number of objects is large, it is a good idea to limit the batch size and wait for CRR to complete between invocations.

Examples:

MAX_UPDATES=10000

This limits the number of updates to 10,000 objects, which requeues a maximum of 10,000 objects to replicate before the script stops.

MAX_SCANNED=10000

This limits the number of scanned objects to 10,000 objects before the script stops.

KEY_MARKER

Set to resume from where an earlier invocation stopped (see MAX_UPDATES, MAX_SCANNED).

Example:

KEY_MARKER="some/key"

VERSION_ID_MARKER

Set to resume from where an earlier invocation stopped (see MAX_UPDATES).

Example:

VERSION_ID_MARKER="123456789"

Example use cases

CRR existing objects after setting a replication policy for the first time

For this use case, it's not necessary to pass any extra environment variable, because the default behavior is to process objects without a replication status attached.

To avoid requeuing too many entries at once, pass this value:

export MAX_UPDATES=10000

Re-queue objects stuck in PENDING state

If Kafka has dropped replication entries, leaving objects stuck in a PENDING state without being replicated, pass the following extra environment variables to reprocess them:

export TARGET_REPLICATION_STATUS=PENDING
export MAX_UPDATES=10000

Warning: This may cause replication of objects already in the Kafka queue to repeat. To avoid this, set the backbeat consumer offsets of "backbeat-replication" Kafka topic to the latest topic offsets before launching the script, to skip over the existing consumer log.

Replicate entries that failed a previous replication

If entries have permanently failed to replicate with a FAILED replication status and were lost in the failed CRR API, it's still possible to re-attempt replication later with the following extra environment variables:

export TARGET_REPLICATION_STATUS=FAILED
export MAX_UPDATES=10000

Re-sync a primary site completely to a new DR site

To re-sync objects to a new DR site (for example, when the original DR site is lost) force a new replication of all original objects with the following environment variables (after setting the proper replication configuration to the DR site bucket):

export TARGET_REPLICATION_STATUS=NEW,PENDING,COMPLETED,FAILED
export MAX_UPDATES=10000

Re-sync a DR site back to the primary site

When objects have been lost from the primary site you can re-sync objects from the DR site to the primary site by re-syncing the objects that have a REPLICA status with the following environment variables (after setting the proper replication configuration from the DR bucket to the primary bucket):

export TARGET_REPLICATION_STATUS=REPLICA
export MAX_UPDATES=10000

Empty a versioned bucket

This script deletes all versions of objects in the bucket, including delete markers, and aborts any ongoing multipart uploads to prepare the bucket for deletion.

Note: This deletes the data associated with objects and is not recoverable

node cleanupBuckets.js testbucket1,testbucket2

List objects that failed replication

This script prints the list of objects that failed replication to stdout, following a comma-separated list of buckets. Run the command as

node listFailedObjects testbucket1,testbucket2

Verify existence of sproxyd keys

This script verifies that :

  1. all sproxyd keys referenced by objects in S3 buckets exist on the RING
  2. sproxyd keys are unique across versions of the same object
  3. object metadata is not an empty JSON object '{}'

It can help to identify objects affected by the S3C-1959 bug (1), or one of S3C-2731, S3C-3778 (2), S3C-5987 (3).

The script can also be used to generate block digests from the listing results as seen by the leader, for the purpose of finding discrepancies between raft members, that can be caused by bugs like S3C-5739.

Usage

node verifyBucketSproxydKeys.js

Mandatory environment variables:

  • BUCKETD_HOSTPORT: ip:port of bucketd endpoint

  • SPROXYD_HOSTPORT: ip:port of sproxyd endpoint

  • One of:

    • BUCKETS: comma-separated list of buckets to scan
    • Example : BUCKETS=bucket1,bucket2,bucket3

  • or:

    • RAFT_SESSIONS: comma-separated list of raft sessions to scan
    • Example : 
      docker run --net=host \
-e 'BUCKETD_HOSTPORT=127.0.0.1:9000' \
-e 'SPROXYD_HOSTPORT=127.0.0.1:8181' \
-e 'RAFT_SESSIONS=1' \
ghcr.io/scality/s3utils:1.12.5 \
node verifyBucketSproxydKeys.js

  • or:

    • KEYS_FROM_STDIN: reads objects to scan from stdin if this environment variable is set, where the input is a stream of JSON objects, each of the form: {"bucket":"bucketname","key":"objectkey\u0000objectversion"} - note: if \u0000objectversion is not present, it checks the master key
    • Example : 
      cat raft_session_1_output.txt | docker run -i --net=host --rm \
-e 'SPROXYD_HOSTPORT=127.0.0.1:8181' \
-e 'BUCKETD_HOSTPORT=127.0.0.1:9000' \
-e 'KEYS_FROM_STDIN=1' \
ghcr.io/scality/s3utils:1.12.5 \
node verifyBucketSproxydKeys.js | tee ring_scan_raft_session_1.txt

Optional environment variables:

  • WORKERS: concurrency value for sproxyd requests (default 100)

  • FROM_URL: URL from which to resume scanning ("s3://bucket[/key]")

  • VERBOSE: set to 1 for more verbose output (shows one line for every sproxyd key)

  • LOG_PROGRESS_INTERVAL: interval in seconds between progress update log lines (default 10)

  • LISTING_LIMIT: number of keys to list per listing request (default 1000)

  • MPU_ONLY: only scan objects uploaded with multipart upload method

  • NO_MISSING_KEY_CHECK: do not check for existence of sproxyd keys, for a performance benefit - other checks like duplicate keys and missing metadata are still done

  • LISTING_DIGESTS_OUTPUT_DIR: output listing digests into the specified directory (in the LevelDB format)

  • LISTING_DIGESTS_BLOCK_SIZE: number of keys in each listing digest block (default 1000)

Output

The output of the script consists of JSON log lines. The most important ones are described below.

Info

progress update

This log message is reported at regular intervals, every 10 seconds by default unless LOG_PROGRESS_INTERVAL environment variable is defined. It shows statistics about the scan in progress.

Logged fields:

  • scanned: number of objects scanned

  • skipped: number of objects skipped (in case MPU_ONLY is set)

  • haveMissingKeys: number of object versions found with at least one missing sproxyd key

  • haveDupKeys: number of object versions found with at least one sproxyd key shared with another object version

  • haveEmptyMetadata: number of objects found with an empty metadata blob i.e. {}

  • url: current URL s3://bucket[/object] for the current listing iteration, can be used to resume a scan from this point passing FROM_URL environment variable.

Example:

{"name":"s3utils:verifyBucketSproxydKeys","time":1585700528238,"skipped":0,"scanned":115833,
"haveMissingKeys":2,"haveDupKeys":0,"url":"s3://some-bucket/object1234","level":"info",
"message":"progress update","hostname":"e923ec732b42","pid":67}

Issue reports

sproxyd check reported missing key

This log message is reported when a missing sproxyd key has been found in an object. The message may appear more than once for the same object in case multiple sproxyd keys are missing.

Logged fields:

  • objectUrl: URL of the affected object version: s3://bucket/object[%00UrlEncodedVersionId]

  • sproxydKey: sproxyd key that is missing

duplicate sproxyd key found

This message is reported when two versions (possibly the same) are sharing an identical, duplicated sproxyd key.

It represents a data loss risk, since data loss would occur when deleting one of the versions with a duplicate key, causing the other version to refer to deleted data in its location array.

Logged fields:

  • objectUrl: URL of the first affected object version with an sproxyd key shared with the second affected object: s3://bucket/object[%00UrlEncodedVersionId]

  • objectUrl2: URL of the second affected object version with an sproxyd key shared with the first affected object: s3://bucket/object[%00UrlEncodedVersionId]

  • sproxydKey: sproxyd key that is duplicated between the two object versions

object with empty metadata found

This message is reported when an object that has an empty metadata blob as its value is found.

Those objects may have been created during an upgrade scenario where connectors were upgraded to a S3C version >= 7.4.10.2, but metadata stateful nodes were not yet upgraded, resulting in the bug described by S3C-5987.

Such objects can normally be safely deleted because they are generated only when an "AbortMultipartUpload" operation is executed on the object, meaning that the object was the target of an MPU in progress that was aborted, and it was the last state of this object (otherwise it would have been updated by non-empty metadata).

Note: even on versioned buckets, the affected keys are always non-versioned.

Logged fields:

  • objectUrl: URL of the affected object: s3://bucket/object

How to delete all such objects:

Assuming verifyBucketSproxydKeys.log contains the output of the previous run of the verify script, this shell command can be used to delete the objects with empty metadata:

JQ_SCRIPT='
select(.message == "object with empty metadata found") |
.objectUrl |
sub("s3://";"http://localhost:9000/default/bucket/")'
jq -r "${JQ_SCRIPT}" verifyBucketSproxydKeys.log | while read url; do
    [ "$(curl -s "${url}")" = '{}' ] && {
        echo "deleting ${url}";
        curl -XDELETE "${url}" || echo "failed to delete ${url}";
    } || true
done

Compare follower's databases against leader view

The CompareRaftMembers/followerDiff tool compares Metadata leveldb databases on the repd follower on which it is run against the leader's view, and outputs the differences to the file path given as the DIFF_OUTPUT_FILE environment variable.

In this file, it outputs each key that differs as line-separated JSON entries, where each entry can be one of:

  • [{ key, value }, null]: this key is present on this follower but not on the leader

  • [null, { key, value }]: this key is not present on this follower but is present on the leader

  • [{ key, value: "{value1}" }, { key, value: "{value2}" }]: this key has a different value between this follower and the leader: "value1" is the value seen on the follower and "value2" the value seen on the leader.

It is possible and recommended to speed-up the comparison by providing a pre-computed digests database via the LISTING_DIGESTS_INPUT_DIR environment variable, so that ranges of keys that match the digests database do not have to be checked by querying the leader. The pre-computed digests database can be generated via a run of "verifyBucketSproxydKeys" script, providing it the LISTING_DIGESTS_OUTPUT_DIR environment variable.

Usage

node followerDiff.js

Mandatory environment variables

  • BUCKETD_HOSTPORT: ip:port of bucketd endpoint

  • DATABASES: space-separated list of databases to scan

  • DIFF_OUTPUT_FILE: file path where diff output will be stored

Optional environment variables

  • LISTING_DIGESTS_INPUT_DIR: read listing digests from the specified LevelDB database

  • PARALLEL_SCANS: number of databases to scan in parallel (default 4)

  • EXCLUDE_FROM_CSEQS: mapping of raft sessions to filter on, where keys are raft session IDs and values are the cseq value for that raft session. Filtering will be based on all oplog records more recent than the given "cseq" for the raft session. Input diff entries not belonging to one of the declared raft sessions are discarded from the output. The value must be in the following JSON format:

    • {"rsId":cseq[,"rsId":cseq...]}

    • Example: {"1":1234,"4":4567,"6":6789}

    • This configuration would cause diff entries of which bucket/key appear in one of the following to be discarded from the output:

      • oplog of raft session 1 after cseq=1234

      • or oplog of raft session 4 after cseq=4567

      • or oplog of raft session 6 after cseq=6789

      • or any other raft session's oplog at any cseq

Scan Procedure with Stable Leader

Note: you may use this procedure when leaders are stable, i.e. leader changes are rare in normal circumstances. When leaders are unstable, use the unstable leader procedure instead.

In order to scan a set of raft sessions to detect discrepancies between raft members, you may use the following procedure involving the two scan tools verifyBucketSproxydKeys and followerDiff.

Make sure that no leader change occurs from the beginning to the end of running this procedure.

The general idea is to compare each follower's view in turn against the current leader's view, without disturbing the leader in order to avoid downtime.

To speed up the comparisons, a digests database is first generated from the current leader's view of all raft sessions, which is then used by individual comparisons on followers to skim quickly over what hasn't changed without sending requests to the leader.

  • Step Pre-check is to be run on the supervisor

  • Steps Prep-1 to Prep-3 are to be run once, on a host having access to bucketd as well as having direct SSH access to stateful hosts (e.g. the supervisor)

  • Step Scan-1 to Scan-4 are to be run repeatedly for each active stateful host, so five times on most systems

  • Step Results-1 is to be run once at the end of the scan, to gather results

Note: If running on RH8 or Rocky Linux 8, you may adapt the given docker commands with crictl instead.

Pre-check: followers must be up-to-date

On each raft session, check that all followers are up-to-date with the leader.

You may use the following command on the supervisor to check this:

./ansible-playbook -i env/s3config/inventory tooling-playbooks/check-status-metadata.yml

Results are gathered in /tmp/results.txt: for all raft sessions, make sure that all committed values across active md[x]-cluster[y] members are within less than 100 entries to each other. Ignore values of wsb[x]-cluster[y] members.

Step Prep-1: Generate the digests database on one metadata server

Start by scanning all raft sessions using verifyBucketSproxydKeys, to generate a digests database in order to have a more efficient scan of followers later on, that does not require an entire listing of the leader again.

Provide the environment variables LISTING_DIGESTS_OUTPUT_DIR and enable NO_MISSING_KEY_CHECK to speed up the listing.

If the command is restarted, e.g. after an error or manual abortion, make sure to delete the ${DIGESTS_PATH} directory before running the command again to generate a clean and efficient database.

Example:

DIGESTS_PATH=~/followerDiff-digests-$(date -I);
docker run \
--net=host \
--rm \
-e 'BUCKETD_HOSTPORT=127.0.0.1:9000' \
-e 'RAFT_SESSIONS=1,2,3,4,5,6,7,8' \
-e 'LISTING_DIGESTS_OUTPUT_DIR=/digests' \
-v "${DIGESTS_PATH}:/digests" \
-e 'NO_MISSING_KEY_CHECK=1' \
ghcr.io/scality/s3utils:1.13.23 \
node verifyBucketSproxydKeys.js \
| tee -a verifyBucketSproxydKeys.log

Step Prep-2: Copy the digests database to each active metadata stateful server

Example:

USER=root
for HOST in storage-1 storage-2 storage-3 storage-4 storage-5; do
    scp -r ${DIGESTS_PATH} ${USER}@${HOST}:followerDiff-digests
done

Step Prep-3: Save the list of raft sessions to scan on each stateful server

The following script computes the list of raft sessions to scan for each stateful server, and outputs the result as a series of ssh commands to execute to save the list on each active stateful. It requires the 'jq' command to be installed where it is executed.

It starts with all listed raft sessions for each active stateful, but removes the raft sessions on each active stateful on which it is the leader.

You may save the following script to a file before executing it with "sh".

#!/bin/sh

### Config

BUCKETD_ENDPOINT=localhost:9000
USER=root
RS_LIST="1 2 3 4 5 6 7 8"


### Script

JQ_SCRIPT_EACH='
.leader.host as $leader
| .connected[]
| select(.host != $leader)
| { host, $rs }'

JQ_SCRIPT_ALL='
group_by(.host)
| .[]
| .[0].host as $host
| "echo \(map(.rs) | join(" ")) > /tmp/rs-to-scan"
| @sh "ssh -n \($user)@\($host) \(.)"'

for RS in $RS_LIST; do
    curl -s http://${BUCKETD_ENDPOINT}/_/raft_sessions/${RS}/info \
    | jq --arg rs ${RS} "${JQ_SCRIPT_EACH}";
done | jq -rs --arg user ${USER} "${JQ_SCRIPT_ALL}"

Executing this script gives on the standard output, the list of ssh commands to run to save the list of raft sessions to scan in a temporary file on each active stateful member, for example:

ssh -n 'root'@'storage-1' 'echo 3 4 5 7 > /tmp/rs-to-scan'
ssh -n 'root'@'storage-2' 'echo 1 2 3 5 6 7 8 > /tmp/rs-to-scan'
ssh -n 'root'@'storage-3' 'echo 1 2 3 4 5 6 8 > /tmp/rs-to-scan'
ssh -n 'root'@'storage-4' 'echo 1 2 3 4 5 6 7 8 > /tmp/rs-to-scan'
ssh -n 'root'@'storage-5' 'echo 1 2 4 6 7 8 > /tmp/rs-to-scan'

Check that what you get looks like the above, then execute those commands on the shell.

Example, assuming you saved the above script as output-rs-to-scan-commands.sh:

# sh output-rs-to-scan-commands.sh
ssh -n 'root'@'storage-1' 'echo 3 4 5 7 > /tmp/rs-to-scan'
ssh -n 'root'@'storage-2' 'echo 1 2 3 5 6 7 8 > /tmp/rs-to-scan'
ssh -n 'root'@'storage-3' 'echo 1 2 3 4 5 6 8 > /tmp/rs-to-scan'
ssh -n 'root'@'storage-4' 'echo 1 2 3 4 5 6 7 8 > /tmp/rs-to-scan'
ssh -n 'root'@'storage-5' 'echo 1 2 4 6 7 8 > /tmp/rs-to-scan'
# sh output-rs-to-scan-commands.sh | bash -x

The following steps are to be run for each active metadata stateful node.

Step Scan-1: SSH to next stateful host to scan

Example:

ssh root@storage-1

Step Scan-2: Gather current cseq of each raft session to scan

This step gathers the current cseq value of the follower repds for all raft sessions to be scanned. The result will be passed on to the followerDiff command in order to remove false positives due to live changes while the script is running.

for RS in $(cat /tmp/rs-to-scan); do
    for REPD_CONF in ${ENV_HOST_DATA}/scality-metadata-bucket/conf/repd_*.json; do
        REPD_RS=$(jq .raftSessionId ${REPD_CONF})
        if [ $RS = $REPD_RS ]; then
            REPD_ADMIN_PORT=$(jq .adminPort ${REPD_CONF})
            echo -n '{"'${RS}'":'
            curl -s "http://localhost:${REPD_ADMIN_PORT}/_/raft/state"
            echo '}'
        fi
    done
done \
| jq -s 'reduce .[] as $item ({}; . + ($item | map_values(.committed)))' \
| tee /tmp/rs-cseqs.json

This command should show a result resembling this, mapping raft session numbers to latest cseq values, and storing it in /tmp/rs-cseqs.json:

{
  "3": 17074,
  "4": 11121,
  "5": 15666,
  "7": 169677
}

Step Scan-3: Stop all follower repd processes

In order to be able to scan the databases, follower repd processes must be stopped with the following command:

for RS in $(cat /tmp/rs-to-scan); do
    for REPD_CONF in ${ENV_HOST_DATA}/scality-metadata-bucket/conf/repd_*.json; do
        REPD_RS=$(jq .raftSessionId ${REPD_CONF})
        if [ $RS = $REPD_RS ]; then
            REPD_CONF_FILE=$(basename ${REPD_CONF})
            REPD_NAME=${REPD_CONF_FILE%%.json}
            docker exec -u root scality-metadata-bucket-repd \
            supervisorctl -c /conf/supervisor/supervisord-repd.conf stop repd:${REPD_NAME}
        fi
    done
done

In case errors occur, restart the command.

Step Scan-4: Scan local databases

Run the "CompareRaftMembers/followerDiff" tool to generate a line-separated JSON output file showing all differences found between the leader and the local metadata databases, for raft sessions where the repd process is a follower.

Note that the container must mount all metadata databases mountpoints in order to have access to all databases.

Example command:

DATABASE_VOLUME_MOUNTS=$(docker inspect scality-metadata-bucket-repd \
| jq -r '.[0].Mounts | map(select(.Source | contains("scality-metadata-databases-bucket")) | "-v \(.Source):\(.Destination)") | .[]')
DATABASE_MASTER_MOUNTPOINT=$(docker inspect scality-metadata-bucket-repd \
| jq -r '.[0].Mounts | map(select(.Source | contains("scality-metadata-databases-bucket") and (contains("/ssd01/") or contains("/ssd1/"))) | .Destination) | .[]')

mkdir -p scan-results
docker run --net=host --rm \
  -e 'BUCKETD_HOSTPORT=localhost:9000' \
  ${DATABASE_VOLUME_MOUNTS} \
  -e "DATABASES_GLOB=$(cat /tmp/rs-to-scan | tr -d '\n' | xargs -d' ' -IRS echo ${DATABASE_MASTER_MOUNTPOINT}/RS/0/*)" \
  -v "${PWD}/followerDiff-digests:/digests" \
  -e "LISTING_DIGESTS_INPUT_DIR=/digests" \
  -v "${PWD}/scan-results:/scan-results" \
  -e "DIFF_OUTPUT_FILE=/scan-results/scan-results.json" \
  -e "EXCLUDE_FROM_CSEQS=$(cat /tmp/rs-cseqs.json)" \
  ghcr.io/scality/s3utils:1.13.23 \
  bash -c 'DATABASES=$(echo $DATABASES_GLOB) node CompareRaftMembers/followerDiff' \
| tee -a followerDiff.log

Note: the tool will refuse to override an existing diff output file. If you need to re-run the command, first delete the output file or give another path as DIFF_OUTPUT_FILE.

Step Scan-5: Restart all stopped repd processes

for RS in $(cat /tmp/rs-to-scan); do
    for REPD_CONF in ${ENV_HOST_DATA}/scality-metadata-bucket/conf/repd_*.json; do
        REPD_RS=$(jq .raftSessionId ${REPD_CONF})
        if [ $RS = $REPD_RS ]; then
            REPD_CONF_FILE=$(basename ${REPD_CONF})
            REPD_NAME=${REPD_CONF_FILE%%.json}
            docker exec -u root scality-metadata-bucket-repd \
            supervisorctl -c /conf/supervisor/supervisord-repd.conf start repd:${REPD_NAME}
        fi
    done
done

In case errors occur, restart the command.

If there are more active stateful hosts to scan, continue with Step Scan-1 on the next active stateful host.

Step Results-1: Gather results

Finally, once all scans have been done on all active stateful hosts, gather all diff results for later analysis or repair.

For example:

for HOST in storage-1 storage-2 storage-3 storage-4 storage-5; do
    scp -r ${USER}@${HOST}:scan-results ./scan-results.${HOST}
done

At this point, each results file contains a newline-separated set of JSON entries describing each difference found on each of the servers, for the databases used by repd processes acting as followers in the Raft protocol. This output is meant to be passed on to the repair script to actually repair the divergences. More details on the output format can be found in the tool's description summary.

Caveats

  • The more up-to-date the digests database is, the more efficient the subsequent stateful host scans are. This means that it's better not to wait too long between the digests database generation and each host scan, in order to reduce the number of updates that occurred in-between, therefore limiting the number of new listings that have to be done on the leader.

  • It is possible to generate the digests database in multiple steps if need be, however in such case it's best not to re-scan the same range of keys multiple times outputting to the same digests database, the risk being to render that range of digests much less efficient to optimize the comparisons, or even useless. If an update of the digests database is wanted, it's best to re-create a new one from scratch (e.g. by deleting the old digests directory first).

  • The current version of the script does not work with Metadata bucket format v1, only with v0. If the need arises, it could be made to work with v1 with some more work.

  • The current version of the script does not support internal TLS feature

Compare two followers database sets

The CompareRaftMembers/compareFollowerDbs tool compares two sets of Metadata leveldb databases, belonging to two different Metadata nodes, and outputs the differences to the file path given as the DIFF_OUTPUT_FILE environment variable.

In this file, it outputs each key that differs as line-separated JSON entries, where each entry can be one of:

  • [{ key, value }, null]: this key is present on the follower #1 but not on follower #2

  • [null, { key, value }]: this key is not present on follower #1 but is present on follower #2

  • [{ key, value: "{value1}" }, { key, value: "{value2}" }]: this key has a different value between follower #1 and follower #2: "value1" is the value seen on follower #1 and "value2" the value seen on follower #2.

Usage

node compareFollowerDbs.js

Mandatory environment variables

  • DATABASES1: space-separated list of databases of follower #1 to compare against follower #2

  • DATABASES2: space-separated list of databases of follower #2 to compare against follower #1

  • DIFF_OUTPUT_FILE: file path where diff output will be stored

Optional environment variables

  • PARALLEL_SCANS: number of databases to scan in parallel (default 4)

  • EXCLUDE_FROM_CSEQS: mapping of raft sessions to filter on, where keys are raft session IDs and values are the cseq value for that raft session. Filtering will be based on all oplog records more recent than the given "cseq" for the raft session. Input diff entries not belonging to one of the declared raft sessions are discarded from the output. The value must be in the following JSON format:

    • {"rsId":cseq[,"rsId":cseq...]}

    • Example: {"1":1234,"4":4567,"6":6789}

    • This configuration would cause diff entries of which bucket/key appear in one of the following to be discarded from the output:

      • oplog of raft session 1 after cseq=1234

      • or oplog of raft session 4 after cseq=4567

      • or oplog of raft session 6 after cseq=6789

      • or any other raft session's oplog at any cseq

  • BUCKETD_HOSTPORT: ip:port of bucketd endpoint, needed when EXCLUDE_FROM_CSEQS is set (in order to read the Raft oplog)

Scan Procedure with Unstable Leader

Note: you may use this procedure when leaders are unstable, i.e. leader changes are frequent or susceptible to happen during the duration of the scan procedure. When leaders are stable, it can be more practical and efficient to use the stable leader procedure instead.

In order to scan a set of raft sessions to detect discrepancies between raft members, you may use the following procedure involving the scan tool compareFollowerDbs.

The general idea is to compare sets of databases belonging to two different followers at a time.

Note: Each repd node (including the current leader) has to be stopped at some point to copy or read its databases set, then restarted. Stopping the leader can have a small temporary impact on the traffic.

Define global configuration values

  • Choose one node on the system that will be the source from which we will copy the set of leveldb databases to a filesystem location that can be shared with other nodes later on. It can be any one of the Metadata active nodes.

    We will call this node SourceNode in the rest of this procedure, and its hostname stored in the environment variable SOURCE_NODE in the provided commands.

  • Choose a list of Raft Sessions to scan

    Decide on which raft sessions will be scanned during this run of the procedure. It may be all data Raft Sessions (usually 8 with IDs 1 through 8), or a subset.

    We will store this set as a space-separated list of raft session IDs, for example 1 2 3 4 5 6 7 8, in the environment variable RAFT_SESSIONS_TO_SCAN in the provided commands.

  • Choose a unique shared filesystem location that:

    • has enough storage space to host Metadata databases of the raft sessions being scanned (from a single node)

    • will be network-accessible from each Metadata node (i.e. contents can be rsync from/to Metadata hosts)

    We will save this remote rsync location as REMOTE_DATABASES_HOST_PATH environment variable in the provided commands, for example root@node-with-storage:/path/to/source-node-dbs

  • Choose a filesystem location on each stateful node that has enough storage space to host Metadata databases of the raft sessions being scanned (from a single node) to receive a copy of the remote databases from SourceNode. It can be for example a location on the RING spinner disks. We will save this path to the environment variable LOCAL_DATABASES_PATH, for example /scality/disk1/source-node-dbs.

  • env_host_data Ansible variable value will be provided as ENV_HOST_DATA environment variable in the provided commands.

The following steps will be run as follows:

  • Step Pre-check is to be run on the supervisor

  • Steps Prep-1 to Prep-5 are to be run on SourceNode only

  • Step Scan-1 to Scan-5 are to be run repeatedly for each active stateful host other than SourceNode, so four times on most systems

  • Step Results-1 is to be run once at the end of the scan, to gather results

Pre-check: followers must be up-to-date

On each raft session, check that all followers are up-to-date with the leader.

You may use the following command on the supervisor to check this:

./ansible-playbook -i env/s3config/inventory tooling-playbooks/check-status-metadata.yml

Results are gathered in /tmp/results.txt: for all raft sessions, make sure that all committed values across active md[x]-cluster[y] members are within less than 100 entries to each other. Ignore values of wsb[x]-cluster[y] members.

Step Prep-1: Gather current cseq of each raft session to scan

This step gathers the current cseq value of the SourceNode repd for all raft sessions to be scanned. The result will be passed on to the compareFollowerDbs command in order to remove false positives due to live changes while the procedure is executed.

for RS in ${RAFT_SESSIONS_TO_SCAN}; do
    for REPD_CONF in ${ENV_HOST_DATA}/scality-metadata-bucket/conf/repd_*.json; do
        REPD_RS=$(jq .raftSessionId ${REPD_CONF})
        if [ $RS = $REPD_RS ]; then
            REPD_ADMIN_PORT=$(jq .adminPort ${REPD_CONF})
            echo -n '{"'${RS}'":'
            curl -s "http://localhost:${REPD_ADMIN_PORT}/_/raft/state"
            echo '}'
        fi
    done
done \
| jq -s 'reduce .[] as $item ({}; . + ($item | map_values(.committed)))' \
| tee /tmp/rs-cseqs.json

This command should show a result resembling this, mapping raft session numbers to latest cseq values, and storing it in /tmp/rs-cseqs.json:

{
  "1": 17074,
  "2": 11121,
  "3": 15666,
  "4": 169677
}

Step Prep-2: Copy cseqs file to other nodes

Copy the file just created on SourceNode to each other node that will be scanned.

Assuming SourceNode is storage-1 and there are 5 nodes to scan, this command would copy the file to each of the other 4 nodes:

for NODE in storage-2 storage-3 storage-4 storage-5; do
    scp /tmp/rs-cseqs.json root@${NODE}:/tmp/
done

Step Prep-3: Stop repd processes

In order to be able to copy the databases, repd processes for the corresponding raft sessions to scan must be stopped.

It can be done with the following script:

for RS in ${RAFT_SESSIONS_TO_SCAN}; do
    for REPD_CONF in ${ENV_HOST_DATA}/scality-metadata-bucket/conf/repd_*.json; do
        REPD_RS=$(jq .raftSessionId ${REPD_CONF})
        if [ $RS = $REPD_RS ]; then
            REPD_CONF_FILE=$(basename ${REPD_CONF})
            REPD_NAME=${REPD_CONF_FILE%%.json}
            docker exec -u root scality-metadata-bucket-repd \
            supervisorctl -c /conf/supervisor/supervisord-repd.conf stop repd:${REPD_NAME}
        fi
    done
done

In case errors occur, restart the command.

Step Prep-4: Copy databases to the remote location

Copy the set of LevelDB databases hosting Metadata for the raft sessions in RAFT_SESSIONS_TO_SCAN, to the remote filesystem location defined as REMOTE_DATABASES_HOST_PATH.

Assuming SourceNode has SSH access to the remote database host, it can be done with the following script:

for RS in ${RAFT_SESSIONS_TO_SCAN}; do
    echo "copying databases for raft session ${RS} to ${REMOTE_DATABASES_HOST_PATH}"
    rsync -a ${ENV_HOST_DATA}/scality-metadata-databases-bucket/${RS} ${REMOTE_DATABASES_HOST_PATH}/
done

In case errors occur, restart the command.

Step Prep-5: Restart repd processes

It can be done with the following script:

for RS in ${RAFT_SESSIONS_TO_SCAN}; do
    for REPD_CONF in ${ENV_HOST_DATA}/scality-metadata-bucket/conf/repd_*.json; do
        REPD_RS=$(jq .raftSessionId ${REPD_CONF})
        if [ $RS = $REPD_RS ]; then
            REPD_CONF_FILE=$(basename ${REPD_CONF})
            REPD_NAME=${REPD_CONF_FILE%%.json}
            docker exec -u root scality-metadata-bucket-repd \
            supervisorctl -c /conf/supervisor/supervisord-repd.conf start repd:${REPD_NAME}
        fi
    done
done

In case errors occur, restart the command.

Important: Run the following Scan steps on each stateful node except SourceNode.

Step Scan-1: SSH to next stateful host to scan

SSH to the next stateful host (excluding SourceNode) where the next scan will run to compare against SourceNode.

Example:

ssh root@storage-2

Step Scan-2: Copy remote databases locally

Copy the set of LevelDB databases previously saved from SourceNode to the remote location locally.

It can be done with the following script:

rsync -a ${REMOTE_DATABASES_HOST_PATH}/ ${LOCAL_DATABASES_PATH}/

Step Scan-3: Stop repd processes

In order to be able to scan the databases, repd processes must be stopped with the following command:

for RS in ${RAFT_SESSIONS_TO_SCAN}; do
    for REPD_CONF in ${ENV_HOST_DATA}/scality-metadata-bucket/conf/repd_*.json; do
        REPD_RS=$(jq .raftSessionId ${REPD_CONF})
        if [ $RS = $REPD_RS ]; then
            REPD_CONF_FILE=$(basename ${REPD_CONF})
            REPD_NAME=${REPD_CONF_FILE%%.json}
            docker exec -u root scality-metadata-bucket-repd \
            supervisorctl -c /conf/supervisor/supervisord-repd.conf stop repd:${REPD_NAME}
        fi
    done
done

In case errors occur, restart the command.

Step Scan-4: Scan local databases

Run the "CompareRaftMembers/compareFollowerDbs" tool to generate a line-separated JSON output file showing all differences found between this node's databases and the SourceNode's databases, for the chosen list of raft sessions to scan.

Note that the container must mount all metadata databases mountpoints in order to have access to all local databases, as well as mount LOCAL_DATABASES_PATH.

Example command:

MY_DATABASE_VOLUME_MOUNTS=$(docker inspect scality-metadata-bucket-repd \
| jq -r '.[0].Mounts | map(select(.Source | contains("scality-metadata-databases-bucket")) | "-v \(.Source):\(.Destination)") | .[]')
DATABASE_MASTER_MOUNTPOINT=$(docker inspect scality-metadata-bucket-repd \
| jq -r '.[0].Mounts | map(select(.Source | contains("scality-metadata-databases-bucket") and (contains("/ssd01/") or contains("/ssd1/"))) | .Destination) | .[]')

mkdir -p scan-results
docker run --net=host --rm \
  -e 'BUCKETD_HOSTPORT=localhost:9000' \
  ${MY_DATABASE_VOLUME_MOUNTS} \
  -v ${LOCAL_DATABASES_PATH}:/databases2 \
  -e "DATABASES1_GLOB=$(echo ${RAFT_SESSIONS_TO_SCAN} | tr -d '\n' | xargs -d' ' -IRS echo ${DATABASE_MASTER_MOUNTPOINT}/RS/0/*)" \
  -e "DATABASES2_GLOB=$(echo ${RAFT_SESSIONS_TO_SCAN} | tr -d '\n' | xargs -d' ' -IRS echo /databases2/RS/0/*)" \
  -v "${PWD}/scan-results:/scan-results" \
  -e "DIFF_OUTPUT_FILE=/scan-results/scan-results.json" \
  -e "EXCLUDE_FROM_CSEQS=$(cat /tmp/rs-cseqs.json)" \
  ghcr.io/scality/s3utils:1.13.23 \
  bash -c 'DATABASES1=$(echo $DATABASES1_GLOB) DATABASES2=$(echo $DATABASES2_GLOB) node CompareRaftMembers/compareFollowerDbs' \
| tee -a compareFollowerDbs.log

Note: the tool will refuse to override an existing diff output file. If you need to re-run the command, first delete the output file or give another path as DIFF_OUTPUT_FILE.

Step Scan-5: Restart stopped repd processes

for RS in ${RAFT_SESSIONS_TO_SCAN}; do
    for REPD_CONF in ${ENV_HOST_DATA}/scality-metadata-bucket/conf/repd_*.json; do
        REPD_RS=$(jq .raftSessionId ${REPD_CONF})
        if [ $RS = $REPD_RS ]; then
            REPD_CONF_FILE=$(basename ${REPD_CONF})
            REPD_NAME=${REPD_CONF_FILE%%.json}
            docker exec -u root scality-metadata-bucket-repd \
            supervisorctl -c /conf/supervisor/supervisord-repd.conf start repd:${REPD_NAME}
        fi
    done
done

In case errors occur, restart the command.

If there are more active stateful hosts to scan, continue with Step Scan-1 on the next active stateful host.

Step Results-1: Gather results

Finally, once all scans have been done on all active stateful hosts except SourceNode, gather all diff results for later analysis or repair.

For example:

for HOST in storage-2 storage-3 storage-4 storage-5; do
    scp -r ${USER}@${HOST}:scan-results ./scan-results.${HOST}
done

At this point, each results file contains a newline-separated set of JSON entries describing each difference found on each of the servers against SourceNode, for the databases used by each repd process on each node. This output is meant to be passed on to the repair script to actually repair the divergences. More details on the output format can be found in the tool's description summary.

Caveats

  • The current version of the script does not work with Metadata bucket format v1, only with v0. If the need arises, it could be made to work with v1 with some more work.

  • The current version of the script does not support internal TLS feature

Repair objects affected by raft divergence

The repair script provided here attempts to automatically repair the object metadata affected by divergences between leaders and followers, to fix consistency while making sure the repaired object is readable. It does not necessarily mean that the repaired version will be the legitimate version from the application point of view, but the tool strives to be non-destructive and best-effort.

This script checks each entry from stdin, corresponding to a line from the output of followerDiff.js, and repairs the object on metadata if deemed safe to do so with a readable version, and if there's no ambiguity on what's the best way to repair.

Usage

    node CompareRaftMembers/repairObjects.js

Mandatory environment variables

  • BUCKETD_HOSTPORT: ip:port of bucketd endpoint

  • SPROXYD_HOSTPORT: ip:port of sproxyd endpoint

Optional environment variables

  • VERBOSE: set to 1 for more verbose output (shows one line for every sproxyd key checked)

  • DRY_RUN: set to 1 to log statuses without attempting to repair anything

Logs

The script outputs logs in JSON. Each object repair analysis or attempt outputs a single log line, for example:

{"name":"s3utils:CompareRaftMembers:repairObjects","time":1658796563031,"bucket":"s3c-5862-nv-0072","key":"test-2022-07-05T21-06-16-960Z/190","repairStatus":"AutoRepair","repairSource":"Leader","repairMaster":false,"level":"info","message":"repaired object metadata successfully","hostname":"jonathan-storage-1","pid":1}

In the above log, the repair status is AutoRepair, which means the script attempted to repair automatically the object metadata, and it succeeded.

The status can be one of:

  • AutoRepair: the object can be repaired automatically (repairSource described the source metadata for the repair, either Leader or Follower).

  • ManualRepair: the object can be repaired but an operator needs to decide on which version to repair from. This is normally when both versions are readable and the script is unable to take an automatic decision. There is currently no support to use objectRepair to help with those cases, but it can be done via the bucketd API.

  • NotRepairable: the object metadata is either absent or unreadable from both leader and follower (the message gives extra detail), so the script did not attempt any repair.

  • UpdatedByClient: this status is given when the repair script detects that the current metadata does not match one of the leader or follower's entry in the current diff entry being processed. This normally means that the application has overwritten the object with new contents since the scan ran. It can then safely be ignored, and the repair script did not attempt to repair anything here.

Additionally, logs may contain such lines, when a check failed on a particular sproxyd key that belongs to one instance of the metadata (whether on leader or follower):

{"name":"s3utils:CompareRaftMembers:repairObjects","time":1658796563000,"bucketdUrl":"http://localhost:9000/default/bucket/s3c-5862-nv-0072/test-2022-07-05T21-06-16-960Z%2F190","sproxydKey":"A553B230F05B0B55C2D090233B280559E730D320","level":"error","message":"sproxyd check reported missing key","hostname":"jonathan-storage-1","pid":1}

Those are expected when metadata divergence exists, they can give some complementary information for diagnosis.

At the end of the run, a single line shows the final count per status (which is also shown during the run every 10 seconds). For example here, we know that there were 2 automatic repairs performed, and one requiring a manual check and repair (which entry can be found from the logs, looking for "ManualRepair" status):

{"name":"s3utils:CompareRaftMembers:repairObjects","time":1658796563064,"countByStatus":{"AutoRepair":2,"AutoRepairError":0,"ManualRepair":1,"NotRepairable":0,"UpdatedByClient":0},"level":"info","message":"completed repair","hostname":"jonathan-storage-1","pid":1}

Repair Procedure

To repair a set of objects with inconsistencies, first make sure that you ran one of the scan procedures (with stable leader or unstable leader) and got results ready from scan-results.[host] directories.

Example Repair Command

Assuming the hosts are named storage-1 through storage-5, this command can be used to launch the repair in dry-run mode first (it's a good practice to get a sense of what the script will attempt to do before executing it for real, looking at the logs first):

cat scan-results.storage-{1..5}/scan-results.jsonl | \
docker run -i --net=host --rm \
  -e "BUCKETD_HOSTPORT=localhost:9000" \
  -e "SPROXYD_HOSTPORT=localhost:8181" \
  -e "DRY_RUN=1" \
  ghcr.io/scality/s3utils:1.13.23 \
  node CompareRaftMembers/repairObjects | tee -a repairObjects.log

If results make sense, it can be executed without the "DRY_RUN" option to execute the possible automatic repairs, if any:

cat scan-results.storage-{1..5}/scan-results.jsonl | \
docker run -i --net=host --rm \
  -e "BUCKETD_HOSTPORT=localhost:9000" \
  -e "SPROXYD_HOSTPORT=localhost:8181" \
  ghcr.io/scality/s3utils:1.13.23 \
  node CompareRaftMembers/repairObjects | tee -a repairObjects.log

Remove delete markers

The removeDeleteMarkers.js script removes delete markers from one or more versioning-suspended bucket(s).

Usage

    node removeDeleteMarkers.js bucket1[,bucket2...]

Mandatory environment variables

  • ENDPOINT: S3 endpoint

  • ACCESS_KEY: S3 account access key

  • SECRET_KEY: S3 account secret key

Optional environment variables:

  • TARGET_PREFIX: only process a specific prefix in the bucket(s)

  • WORKERS: concurrency value for listing / batch delete requests (default 10)

  • LOG_PROGRESS_INTERVAL: interval in seconds between progress update log lines (default 10)

  • LISTING_LIMIT: number of keys to list per listing request (default 1000)

  • KEY_MARKER: resume processing from a specific key

  • VERSION_ID_MARKER: resume processing from a specific version ID

Output

The output of the script consists of JSON log lines.

Info

One log line is output for each delete marker deleted, e.g.:

{"name":"s3utils::removeDeleteMarkers","time":1586304708269,"bucket":"some-bucket","objectKey":"some-key","versionId":"3938343133363935343035383633393939393936524730303120203533342e3432363436322e393035353030","level":"info","message":"delete marker deleted","hostname":"c3f84aa473a2","pid":88}

The script also logs a progress update, every 10 seconds by default:

{"name":"s3utils::removeDeleteMarkers","time":1586304694304,"objectsListed":257000,"deleteMarkersDeleted":0,"deleteMarkersErrors":0,"bucketInProgress":"some-bucket","keyMarker":"some-key-marker","versionIdMarker":"3938343133373030373134393734393939393938524730303120203533342e3430323230362e373139333039","level":"info","message":"progress update","hostname":"c3f84aa473a2","pid":88}

Repair duplicate sproxyd keys

This script repairs object versions that share sproxyd keys with another version, particularly due to bug S3C-2731.

The repair action consists of copying the data location that is duplicated to a new sproxyd key (or set of keys for MPU), and updating the metadata to reflect the new location, resulting in two valid versions with distinct data, though identical in content.

The script does not remove any version even if the duplicate was due to an internal retry in the metadata layer, because either version might be referenced by S3 clients in some cases.

Usage

node repairDuplicateVersions.js

Standard Input

The standard input must be fed with the JSON logs output by the verifyBucketSproxydKeys.js s3utils script. This script only processes the log entries containing the message "duplicate sproxyd key found" and ignores other entries.

Mandatory environment variables

  • OBJECT_REPAIR_BUCKETD_HOSTPORT: ip:port of bucketd endpoint

  • OBJECT_REPAIR_SPROXYD_HOSTPORT: ip:port of sproxyd endpoint

Example

cat /tmp/verifyBucketSproxydKeys.log | docker run -i zenko/s3utils:latest bash -c 'OBJECT_REPAIR_BUCKETD_HOSTPORT=127.0.0.1:9000 OBJECT_REPAIR_SPROXYD_HOSTPORT=127.0.0.1:8181 node repairDuplicateVersions.js' > /tmp/repairDuplicateVersions.log

Caveat

While the script is running, there is a possibility, although slim, that it updates metadata for an object that is being updated at the exact same time by a client application, either through "PutObjectAcl" or "PutObjectTagging" operations which can modify existing versions. In such case, there is a risk that either:

  • the application update will be lost
  • or the script will not repair the item properly

If this looks like a potential risk to a customer running the script, we suggest to disable those operations on the clients during the time the script is running (or alternatively, disable all writes), to avoid any such risk.

Repair object metadata with duplicate 'versionId' field

This script repairs object versions that have a duplicate versionId field in the master key and a missing version key, in particular due to bug S3C-7861 (CRR with incompatible S3C versions between source and target).

The repair action consists of fetching the source metadata from the master key, fixing the JSON versionId field to be the first versionId present in the binary blob (assumed to be the original one, hence correct), then copying it to both the master key and a new version key.

Usage

node repairDuplicateVersionIds.js

Standard Input

The standard input must be fed with the JSON logs output by the verifyBucketSproxydKeys.js s3utils script. This script only processes the log entries containing the message 'object master metadata with duplicate "versionId" field found' and ignores other entries.

Mandatory environment variables

  • OBJECT_REPAIR_BUCKETD_HOSTPORT: ip:port of bucketd endpoint

Example

cat /tmp/verifyBucketSproxydKeys.log | docker run -i ghcr.io/scality/s3utils:1.13.24 bash -c 'OBJECT_REPAIR_BUCKETD_HOSTPORT=127.0.0.1:9000 node repairDuplicateVersionIds.js' > /tmp/repairDuplicateVersionIds.log

Cleanup Noncurrent Versions

This script removes noncurrent versions and current/noncurrent delete markers, either all such objects or older than a specified last-modified date.

Usage

node cleanupNoncurrentVersions.js bucket1[,bucket2...]

Mandatory environment variables

  • S3_ENDPOINT: S3 endpoint URL

  • ACCESS_KEY: S3 account/user access key

  • SECRET_KEY: S3 account/user secret key

Optional environment variables

  • TARGET_PREFIX: cleanup only inside this key prefix in each bucket

  • MAX_LISTED: maximum number of keys listed before exiting (default unlimited)

  • MAX_DELETES: maximum number of keys to delete before exiting (default unlimited)

  • MARKER: marker from which to resume the cleanup, logged at the end of a previous invocation of the script, uses the format:

    MARKER := encodeURI(bucketName)
  "|" encodeURI(key)
  "|" encodeURI(versionId)

  • OLDER_THAN: cleanup only objects which last modified date is older than this, as an ISO date or a number of days, e.g.:

    • setting to "2021-01-09T00:00:00Z" limits the cleanup to objects created or modified before Jan 9th 2021

    • setting to "30 days" limits the cleanup to objects created more than 30 days ago

  • ONLY_DELETED: if set to "1" or "true" or "yes", only remove otherwise eligible noncurrent versions if the object's current version is a delete marker (also removes the delete marker)

  • DELETED_BEFORE: cleanup only objects whose current version is a delete marker older than this date, e.g. setting to "2021-01-09T00:00:00Z" limits the cleanup to objects deleted before Jan 9th 2021. Implies ONLY_DELETED = true

  • HTTPS_CA_PATH: path to a CA certificate bundle used to authentify the S3 endpoint

  • HTTPS_NO_VERIFY: set to 1 to disable S3 endpoint certificate check

  • EXCLUDE_REPLICATING_VERSIONS: if is set to '1,' 'true,' or 'yes,' prevent the deletion of replicating versions

Example

docker run --net=host -ti zenko/s3utils:latest bash -c 'S3_ENDPOINT=https://s3.customer.com ACCESS_KEY=123456 SECRET_KEY=789ABC OLDER_THAN="Jan 14 2021" node cleanupNoncurrentVersions.js target-bucket-1,target-bucket-2' > /tmp/cleanupNoncurrentVersions.log

Count number and total size of current and noncurrent versions in a bucket

The bucketVersionsStats.js script counts the total number of objects and their cumulative size, and shows them separately for current and noncurrent versions.

Usage

    node bucketVersionsStats.js

Mandatory environment variables

  • ENDPOINT: S3 endpoint

  • ACCESS_KEY: S3 account access key

  • SECRET_KEY: S3 account secret key

  • BUCKET: S3 bucket name

Optional environment variables:

  • TARGET_PREFIX: only process a specific prefix in the bucket

  • LISTING_LIMIT: number of keys to list per listing request (default 1000)

  • LOG_PROGRESS_INTERVAL: interval in seconds between progress update log lines (default 10)

  • KEY_MARKER: start counting from a specific key

  • VERSION_ID_MARKER: start counting from a specific version ID

Output

The output of the script consists of JSON log lines.

The script logs a progress update, every 10 seconds by default, and a final summary at the end of execution with the total counts for the bucket.

The total counts may be possibly limited by the TARGET_PREFIX, KEY_MARKER and VERSION_ID_MARKER optional environment variables.

  • stats contains the output statistics:

    • current refers to current versions of objects

    • noncurrent refers to non-current versions of objects

    • total is the sum of current and noncurrent

  • in stats.current and stats.noncurrent:

    • count is the number of objects pertaining to the section

    • size is the cumulative size in bytes of the objects pertaining to the section

Example Progress Update

Note: the JSON output is prettified here for readability, but the script outputs this status on a single line.

{
  "name": "s3utils::bucketVersionsStats",
  "time": 1638823524419,
  "bucket": "example-bucket",
  "stats": {
    "total": {
      "count": 16000,
      "size": 27205000
    },
    "current": {
      "count": 14996,
      "size": 26165000
    },
    "noncurrent": {
      "count": 1004,
      "size": 1040000
    }
  },
  "keyMarker": "test-2021-12-06T20-09-39-712Z/1732",
  "versionIdMarker": "3938333631313738363132353838393939393939524730303120203131302e353533312e3135373439",
  "level": "info",
  "message": "progress update",
  "hostname": "lab-store-0",
  "pid": 717
}

Example Final Summary

Note: the JSON output is prettified here for readability, but the script outputs this status on a single line.

{
  "name": "s3utils::bucketVersionsStats",
  "time": 1638823531629,
  "bucket": "example-bucket",
  "stats": {
    "total": {
      "count": 54267,
      "size": 65472000
    },
    "current": {
      "count": 53263,
      "size": 64432000
    },
    "noncurrent": {
      "count": 1004,
      "size": 1040000
    }
  },
  "level": "info",
  "message": "final summary",
  "hostname": "lab-store-0",
  "pid": 717
}

Verify Replication

The verify replication script checks/compares the replication source with the replication destination. The script produces logs about missing objects on destination and destination objects whose size doesnt match the source size. The script verifies only current version of objects. It relies on standard s3 api's for the source and the destination.

Usage

node VerifyReplication/index.js

Mandatory environment variables

  • SRC_ENDPOINT: replication source s3 endpoint

  • SRC_BUCKET: replication source bucket

  • SRC_ACCESS_KEY: replication source s3 account access key

  • SRC_SECRET_KEY: replication source s3 account secret key

  • DST_BUCKET: replication destination bucket

  • DST_ACCESS_KEY: replication destination s3 account access key

  • DST_SECRET_KEY: replication destination s3 account secret key

Optional environment variables:

  • SRC_BUCKET_PREFIXES: comma separated list of prefixes, listing will be limited to these prefixes

  • SRC_DELIMITER: delimiter used with prefixes in source (default '/')

  • LISTING_LIMIT: number of keys to list per listing request (default 1000)

  • LISTING_WORKERS: number of concurrent workers to handle listing (default 10)

  • LOG_PROGRESS_INTERVAL: interval in seconds between progress update log lines (default 10)

  • BUCKET_MATCH: set to 1 if bucket match is enabled with replication, when not enabled objects will be replicated with keyname pattern sourceBucket/key

  • COMPARE_OBJECT_SIZE: set to 1 to compare source and destination object sizes

  • DST_ENDPOINT: replication destination s3 endpoint (only for storage type aws_s3)

  • DST_STORAGE_TYPE: destination storage type, currently supports only aws_s3

  • DST_REGION: destination s3 region (only for storage type aws_s3, default us-east-1)

  • DST_MD_REQUEST_WORKERS: number of concurrent workers to handle destination metadata requests (default 50)

  • HTTPS_CA_PATH: path to a CA certificate bundle used to authentify the source S3 endpoint

  • HTTPS_NO_VERIFY: set to 1 to disable source S3 endpoint certificate check

  • SHOW_CLIENT_LOGS_IF_AVAILABLE: set to 1 to show storage client logs if available (enable only for debugging as it may produce a lot of log entries), this may be disregarded if the client (depending on storage type) does not support it

  • SKIP_OLDER_THAN: skip replication verification of objects whose last modified date is older than this, set this as an ISO date or a number of days e.g.,

    • setting to "2022-11-30T00:00:00Z" skips the verification of objects created or modified before Nov 30th 2022
    • setting to "30 days" skips the verification of objects created or modified more than 30 days ago

Output

The output of the script consists of JSON log lines.

Info

One log line is output for each missing object,

{
  "name":"s3utils:verifyReplication",
  "time":1670279729106,
  "key":"rna-4/rna33",
  "size":0,"srcLastModified":"2022-12-01T12:27:11.469Z",
  "level":"info",
  "message":"object missing in destination",
  "hostname":"scality.local",
  "pid":24601
}

and for each size mismatched object,

{
  "name":"s3utils:verifyReplication",
  "time":1670279723474,
  "key":"rna-2/34",
  "srcSize":1,
  "dstSize":0,
  "srcLastModified":"2022-12-01T11:31:27.528Z",
  "dstLastModified":"2022-12-01T11:33:42.000Z",
  "level":"info",
  "message":"object size does not match in destination",
  "hostname":"scality.local",
  "pid":24601
}

and for each failed metadata retrieval (after retries),

{
  "name":"s3utils:verifyReplication",
  "time":1670887351694,
  "error":{
    "message":"socket hang up",
    "code":"TimeoutError",
    "time":"2022-12-12T23:22:31.694Z",
    "region":"us-east-1",
    "hostname":"localhost",
    "retryable":true
    },
  "bucket":"src-bucket",
  "key":"pref1/key404",
  "level":"error",
  "message":"error getting metadata",
  "hostname":"scality.local",
  "pid":11398
}

The script also logs a progress update (a summary), every 10 seconds by default,

{
  "name":"s3utils:verifyReplication",
  "time":1670280035601,
  "srcListedCount":492,
  "dstProcessedCount":492,
  "skippedByDate": 123,
  "missingInDstCount":123,
  "sizeMismatchCount":123,
  "replicatedCount":120,
  "dstFailedMdRetrievalsCount":3,
  "dstBucket":"dst-bucket",
  "srcBucket":"src-bucket",
  "prefixFilters":["pref1","pref2"],
  "skipOlderThan": "2022-11-30T00:00:00Z",
  "level":"info",
  "message":"completed replication verification",
  "hostname":"scality.local",
  "pid":24780
}

Description of the properties in the summary/progress update log,

  • srcListedCount - total number of objects listed from source
  • dstProcessedCount - total number of objects checked in destination (includes missing, mismatched & failed)
  • skippedByDate - total number of objects skipped verification because of a date filter SKIP_OLDER_THAN
  • missingInDstCount - total number of objects missing in destination
  • sizeMismatchCount - total number of objects with mismtached size in destination
  • dstFailedMdRetrievalsCount - total number of failed object metadata retrievals (after retries) from destination
  • replicatedCount - total number of successful replications
  • srcBucket - source bucket
  • dstBucket - destination bucket

Example

docker run \
  --net=host \
  -e 'SRC_ENDPOINT=http://source.s3.com:8000' \
  -e 'SRC_BUCKET=src-bucket' \
  -e 'SRC_ACCESS_KEY=src_access_key' \
  -e 'SRC_SECRET_KEY=src_secret_key' \
  -e 'DST_ENDPOINT=http://destination.s3.com:8000' \
  -e 'DST_ACCESS_KEY=dst_access_key' \
  -e 'DST_SECRET_KEY=dst_secret_key' \
  -e 'DST_BUCKET=dst-bucket' \
  -e 'BUCKET_MATCH=1' \
  -e 'COMPARE_OBJECT_SIZE=1' \
  -e 'SRC_BUCKET_PREFIXES=pref1,pref2' \
  -e 'SKIP_OLDER_THAN="2022-11-30T00:00:00Z"' \
  ghcr.io/scality/s3utils:latest \
  node VerifyReplication/index.js

UtapiV2 Service Report

Generates a service level usage report from UtapiV2 in HTML or JSON format.

Usage

Commands must be executed on a stateful server that hosts the target UtapiV2 deployment.

docker run --net=host --entrypoint python3 -v /scality/ssd01/s3/scality-utapi/conf:/conf:ro -v $PWD:/output zenko/s3utils utapi/service-report.py --output /output

Flags

  -c CONFIG, --config CONFIG
                        Specify an alternate config file (default: /conf/config.json)
  -r MAX_RETRIES, --max-retries MAX_RETRIES
                        Max retries before failing a request to an external service (default: 2)
  -p PARALLEL_QUERIES, --parallel-queries PARALLEL_QUERIES
                        Max number of parallel queries to warp 10 (default: 5)
  -j, --json            Output raw reports in json format (default: False)
  -o OUTPUT, --output OUTPUT
                        Write report to this directory (default: Current Directory)
  --debug               Enable debug level logging (default: False)
  --dry-run             Don't do any computation. Only validate and print the configuration. (default: False)

UtapiV2 Service level metrics sidecar

REST API to provide service level reports for UtapiV2

Usage

docker run -d \
  --network=host \
  -e SIDECAR_API_KEY=dev_key_change_me \
  -e SIDECAR_SCALE_FACTOR=1.4 \
  -e SIDECAR_WARP10_NODE="md1-cluster1:[email protected]:4802" \
  scality/s3utils service-level-sidecar/index.js

curl -X POST -H "Authorization: Bearer dev_key_change_me" localhost:24742/api/report | jq
{
  "account": [
    {
      "arn": "arn:aws:iam::081797933446:/test_1669749866/",
      "name": "test_1669749866",
      "obj_count": 25,
      "bytes_stored": 25,
      "bytes_stored_total": 35
    },
    {
      "arn": "arn:aws:iam::024022147664:/test_1669748782/",
      "name": "test_1669748782",
      "obj_count": 25,
      "bytes_stored": 25,
      "bytes_stored_total": 35
    }
  ],
  "bucket": {
    "arn:aws:iam::081797933446:/test_1669749866/": [
      {
        "name": "test3",
        "obj_count": 25,
        "bytes_stored": 25,
        "bytes_stored_total": 35
      }
    ],
    "arn:aws:iam::024022147664:/test_1669748782/": [
      {
        "name": "test",
        "obj_count": 15,
        "bytes_stored": 15,
        "bytes_stored_total": 21
      },
      {
        "name": "test2",
        "obj_count": 10,
        "bytes_stored": 10,
        "bytes_stored_total": 14
      }
    ]
  },
  "service": {
    "obj_count": 50,
    "bytes_stored": 50,
    "bytes_stored_total": 70
  }
}

Authentication

By default a random API key is generated and logged to stdout at startup.

{"name":"s3utils::utapi-service-sidecar","time":1669677797644,"key":"c489cc0f5bec1c757be7aecd7cdd61dc9db5bfbc4e4f7bad","level":"info","message":"using random API key","hostname":"f2988d47e450","pid":1}

A custom static key can be set using the SIDECAR_API_KEY environment variable.

docker run -d \
  --network=host \
  -e SIDECAR_API_KEY=dev_key_change_me \
  scality/s3utils service-level-sidecar/index.js

API requests to /api/report require this API token to be present in the Authorization header.

curl -X POST -H "Authorization: Bearer dev_key_change_me" localhost:24742/api/report

Scale Factor

Scale factor is provided as a floating point number that represents the percentage to scale reported byte values. It is used to account for the overhead of erasure coding on the storage backend. For example, a value of 1.4 will output byte numbers 140% of the calculated value ie 100 bytes will be reported as 140 bytes. If this produces a non-integer result it is rounded up to the next whole number.

It can be set using the SIDECAR_SCALE_FACTOR environment variable.

docker run -d \
  --network=host \
  -e SIDECAR_SCALE_FACTOR=1.4 \
  scality/s3utils service-level-sidecar/index.js

TLS

TLS can be enabled for the api server as well as internal connections to vault and bucketd. To enable TLS for the API server specify the path to the TLS certificate and key using the SIDECAR_TLS_KEY_PATH and SIDECAR_TLS_CERT_PATH environment variables. A CA can optionally be set using SIDECAR_TLS_CA_PATH. Certificates should be mounted into the container using a docker volume.

docker run -d \
  --network=host \
  -v $PWD/certs:/certs \
  -e SIDECAR_TLS_CERT_PATH=/certs/cert.pem \
  -e SIDECAR_TLS_KEY_PATH=/certs/key.pem \
  -e SIDECAR_TLS_CA_PATH=/certs/ca.pem \ // CA path is optional
  scality/s3utils service-level-sidecar/index.js

To enable TLS for internal connections to vault or bucketd set SIDECAR_VAULT_ENABLE_TLS=true or SIDECAR_BUCKETD_ENABLE_TLS=true respectively.

docker run -d \
  --network=host \
  -e SIDECAR_VAULT_ENABLE_TLS=true \
  -e SIDECAR_BUCKETD_ENABLE_TLS=true \
  scality/s3utils service-level-sidecar/index.js

Backend addresses

Vault

Vault is configured using SIDECAR_VAULT_ENDPOINT

docker run -d \
  --network=host \
  -e SIDECAR_VAULT_ENDPOINT=127.0.0.1:8500 \
  scality/s3utils service-level-sidecar/index.js

Bucketd

Bucketd is configured using SIDECAR_BUCKETD_BOOTSTRAP

docker run -d \
  --network=host \
  -e SIDECAR_BUCKETD_BOOTSTRAP=127.0.0.1:9000 \
  scality/s3utils service-level-sidecar/index.js

Warp10

The Warp 10 address is configured using SIDECAR_WARP10_NODE. The Warp 10 nodeId must be included (normally matches ansible inventory name plus port ie md1-cluster1:4802). The format is <nodeId>@<host>:<port>.

docker run -d \
  --network=host \
  -e SIDECAR_WARP10_NODE="md1-cluster1:[email protected]:4802" \
  scality/s3utils service-level-sidecar/index.js

Other Settings

  • Log level can be set using SIDECAR_LOG_LEVEL (defaults to info)
  • The concurrency used to query backend APIs can be set using SIDECAR_CONCURRENCY_LIMIT (defaults to 10)
  • The maximum retries allowed when querying backend APIs can be set using SIDECAR_RETRY_LIMIT (defaults to 5)
  • The IP address to bind can be set using SIDECAR_HOST (default 0.0.0.0)
  • The port to bind can be set using SIDECAR_PORT (default 24742)