Transients Layered Nav Counts problem (WooCommerce)
In the autumn of 2023, we encountered sudden but persistant problems on WooCommerce webshops, resulting in database server crashes and running out of storage space at a pace of about 2GB per minute. It turned out that this was caused by transients for WooCommerce layered navigation counts, and we actually already had problems related to this for over a year already.
What I'm usually looking for
cd wp-content/plugins/woocommerce/src/Internal/ProductAttributesLookup/ vim wp-content/plugins/woocommerce/src/Internal/ProductAttributesLookup/Filterer.php
Context
- This all plays on our new Hetzner server, which is about 5 to 6 times bigger and faster than our previous server
- There are about 40 webshops on this server with about 17,000 products each.
Immediate problems
These 'immediate problems' nicely illustrate how amorphous computer problems can be: It seems like twice the same problem, or like one problem and one underlying cause - How to keep these kinds of things structured? → Try to keep it intuitive, rather than trying to be 'logically rigid':
Problem | Notes |
---|---|
Random database server crashes? | Every couple of days, the MariaDB server crashed, seemingly due to unlikely issues, like memory shortage or something that seemed like compromized data |
Database server crashes due to storage issues | Somehow, /var/logs/php/php.error was growing with about 2GB per minute, crashing the database server when the server run out of storage space
|
Cause
We have the problem described here at GitHub, but with a minor difference:
- The situation described in this article, is the 'old' situation when there was only one transient for all filters, called
_transient_wc_layered_nav_countsr
- To alluviate this problem, Woo decided to split this one transient into separate transients for each filter [1]
- Instead of this one transient, we now have several transients with names like
_transient_wc_layered_nav_counts_pa_<name of the attribute filter>
.
Despite this difference, we have the exact symptomps described in the link above.
Symptoms
What we experience, and that I found back in the article mentioned above:
- There are a couple of transients with names starting with
transient_wc_layered_nav_counts_pa_
that make up about 80% of the size of tablewp_options
- Storage space for table
wp_options
grows with about 2GB per minute, probably because of logging, as these transients are updated constantly - Db server crashes when running out of storage space
- Site performance degredation, even with enough storage space available
- Every couple of days, we have to delete transients - We have been doing this for at least a year already and only now can I connect it with the other issues
- File
/var/log/php/error.log
grows as well with about 2GB/minute. Unknown if this is related
Our symptoms are nicely illustrated by this comment:
I confirm that since WC 3.2.1, I have to purge my MariaDB bin log files every 5 days. With the current traffic I get on my site (which is yet little trafic), approximately every 5 days, this transient string is so huge that it generates up to 8 gigas of log when it's saved to the table.
Another illustration of our symptoms [2]:
We are proactively contacting you because of hyper filling up the storage up to 500GB due to massive transaction logs by the site. The engineering team detected that it’s generating ~150MB/sec of mysql bin logs (transaction logs). From a quick glance, it looks like the wp_options table and a key named _transient_wc_layered_nav_countsr may be the culprit. The key is being overwritten/appended to at a huge rate.
Transients
As an example, these are those transients, copied from a site when the problem was alreay rather under control thanks to scripts to delete stuff:
option_id option_name First characters Size --------- ----------------------------------------- ------------------------- ----- 2168239 _transient_wc_layered_nav_counts_pa_asxxx a:16:{i:0;b:0;s:32:"c5662 0.28M 2168225 _transient_wc_layered_nav_counts_pa_auxxx a:16:{i:0;b:0;s:32:"4c798 0.60M 2168229 _transient_wc_layered_nav_counts_pa_coxxx a:16:{i:0;b:0;s:32:"c3e8d 0.35M 2168221 _transient_wc_layered_nav_counts_pa_depxx a:16:{i:0;b:0;s:32:"56496 0.36M 2168213 _transient_wc_layered_nav_counts_pa_devxx a:16:{i:0;b:0;s:32:"4c23e 0.34M 2168235 _transient_wc_layered_nav_counts_pa_dixxx a:16:{i:0;b:0;s:32:"258a3 0.39M 2168223 _transient_wc_layered_nav_counts_pa_lexxx a:16:{i:0;b:0;s:32:"099f4 0.37M 2168215 _transient_wc_layered_nav_counts_pa_moxxx a:16:{i:0;b:0;s:32:"cc751 0.38M 2168233 _transient_wc_layered_nav_counts_pa_orxxx a:18:{i:0;b:0;s:32:"65b34 10.28M 2168237 _transient_wc_layered_nav_counts_pa_prxxx a:16:{i:0;b:0;s:32:"4ae79 1.86M 2168231 _transient_wc_layered_nav_counts_pa_sprxx a:16:{i:0;b:0;s:32:"c5e23 0.36M 2168217 _transient_wc_layered_nav_counts_pa_subxx a:16:{i:0;b:0;s:32:"52a9d 0.34M 2168219 _transient_wc_layered_nav_counts_pa_widxx a:16:{i:0;b:0;s:32:"7735d 14.90M 2168227 _transient_wc_layered_nav_counts_pa_wirxx a:16:{i:0;b:0;s:32:"a04c7 0.34M Total size: 31.15MB
and this is how their names more-or-less look like, when using e.g., wp transient list
:
wc_layered_nav_counts_pa_dim2-x wc_layered_nav_counts_pa_dim1-x wc_layered_nav_counts_pa_device wc_layered_nav_counts_pa_code wc_layered_nav_counts_pa_automati-x wc_layered_nav_counts_pa_breed-x wc_layered_nav_counts_pa_connector wc_layered_nav_counts_pa_diagram wc_layered_nav_counts_pa_diep-x wc_layered_nav_counts_pa_leng-x wc_layered_nav_counts_pa_model wc_layered_nav_counts_pa_origine-x wc_layered_nav_counts_pa_product-x wc_layered_nav_counts_pa_stro-x wc_layered_nav_counts_pa_submodel wc_layered_nav_counts_pa_ve-x
What seemed to be happening
Each webshop has about 9 to 14 attribute taxonomies + associated filters at the /shop
page. The desciption below is not complete nor consistent: Like always, everything is changing all the time. Especially: While I was measuring stuff, someone else was fixing the problems:
- For each filter, a transient is created (in table
wp_options
). Such transients easily become up to 20 to 40 MB per field - WP-CLI command
wp transient list all
usually doesn't work anymore by that time - These filters are being updated all the time and that's a problem: This field is locked while it is being written to. WP Transients are just bad on their own [3]
- With every table update, the size of the database log for this table increases. That might be why files like
/var/lib/mysql/example_com/wp_options.ibd
were growing so fast, that the overall storage was decreasing with about 2GB per minute - At the same time, table
wp_options
still had only about 1,400 lines. Individual fields were becomming megabytes in size. The total size of this table was measured through Bash commands likeout4=$(mysqldump ${db} wp_options | wc -c)
- Eventually, the content of these fields become so big, that they can't be updated anymore, because table locking exceeds time-outs, generating errors like
WordPress database error Lock wait timeout exceeded; try restarting transaction for query UPDATE wp_options SET option_value = 'a:1...
- This happened on 2023.12.07 on site at_en for a field of 44MB - When transients can't be updated anymore, this will generate errors in
/var/log/php/error.log
. Each error line includes the complete SQL udpate statement, including the complete content of the new value of the field. By this time,/var/log/php/error.log
starts to grow with about 2GB per minute. I don't know if MySQL storage continues to grow at a similar rate at this point - Already for some years, we have a script that deletes the transients once a week, because otherwise layout of some pages get messed up. Only now I realize that this is part of the bigger picture described here.
In the following chapters, let's move from the problem to a solution:
In WooCommerce, layered navigation refers to a filtering system that helps customers refine their search results while browsing a store. It allows shoppers to narrow down product selections based on various attributes or criteria, such as price range, size, color, brand, or any custom product attributes set by the store owner.
Layered navigation typically appears as a sidebar or series of filters on category pages or search results. When customers select specific criteria or attributes, the displayed products dynamically update to show only items that match the chosen filters. This feature enhances the user experience by making it easier for customers to find the products that meet their specific preferences or requirements.
WC Layered navigation counts refers to counting the number of items that are within a filter. Amongst other, this is used by the WC Layered Nav Widget to decide whether or not to hide an item when count is 0.
Maybe this came out of this issue about how slow the query was that does this job [4]. Hence cachching → [5]
Source code
Roy Lindauer
Code that Roy Lindauer refers to:
// We have a query - let's see if cached results of this query already exist. $query_hash = md5( $query ); $cached_counts = (array) get_transient( 'wc_layered_nav_counts' ); if ( ! isset( $cached_counts[ $query_hash ] ) ) { $results = $wpdb->get_results( $query, ARRAY_A ); $counts = array_map( 'absint', wp_list_pluck( $results, 'term_count', 'term_count_id' ) ); $cached_counts[ $query_hash ] = $counts; set_transient( 'wc_layered_nav_counts', $cached_counts, DAY_IN_SECONDS ); }
This was located in includes » widgets » class-wc-widget-layered-nav.php
, but it isn't there anymore.
Nowadays
The only place where some of the keywords from above come together, is in src » Internal » ProductAttributesLookup » Filterer.php
around line 171:
// We have a query - let's see if cached results of this query already exist. $query_hash = md5( $query_sql ); // Maybe store a transient of the count values. $cache = apply_filters( 'woocommerce_layered_nav_count_maybe_cache', true ); if ( true === $cache ) { $cached_counts = (array) get_transient( 'wc_layered_nav_counts_' . sanitize_title( $taxonomy ) ); } else { $cached_counts = array(); } if ( ! isset( $cached_counts[ $query_hash ] ) ) { // phpcs:ignore WordPress.DB.PreparedSQL.NotPrepared $results = $wpdb->get_results( $query_sql, ARRAY_A ); $counts = array_map( 'absint', wp_list_pluck( $results, 'term_count', 'term_count_id' ) ); $cached_counts[ $query_hash ] = $counts; if ( true === $cache ) { set_transient( 'wc_layered_nav_counts_' . sanitize_title( $taxonomy ), $cached_counts, DAY_IN_SECONDS ); } }
More places to search
Places where this stuff might reside, below wp-content » plugins » woocommerce
(some less-likely places, like theme files, are excluded):
* abstracts » abstact-wc-widget.php * includes » class-wc-cache-helper.php * includes » class-wc-query.php * includes » class-wc-template-loader.php * includes » wc-conditional-functions.php * includes » wc-widget-functions.php * includes » widgets » class-wc-widget-layered-nav-filters.php * includes » widgets » class-wc-widget-layered-nav.php * includes » widgets » class-wc-widget-rating-filter.php
Most likely locations:
* includes » class-wc-query.php (query answer Stack Exchange) * includes » widgets » class-wc-widget-layered-nav.php (Roy Lindauer, not there anymore) * src » Internal » ProductAttributesLookup » Filterer.php - Some relevant keywords
Potential solutions
# | Solution | Description | Evaluation |
---|---|---|---|
1 | Use memcach | [6] | Using in-memory caching might be a real solution, rather than suppressing symptoms → See Use object cache below. |
2 | Flush transients through cron when too big | [7]. By default, they hang around for about 1 day, so it size is an issue, they should probably be flushed more often, e.g., once per hour |
|
3 | Upcoming Woo solution | Wait until Woo comes with a 'real' solution, alluded to in the main article | No: Let's create a dependency on something we can't control and that isn't there yet |
4 | Remove transients creation from WooCommerce code |
|
|
5 | Use an object cache | Use an object cache like Memcache or Reddis to store these transients in-memory, rather than in a database [8]. Also: This single record becomes a bottleneck as the field is locked while it is being written to. Redis or Memcache would be a a better solution. WP Transients are just bad on their own. [9] | In our case, these transients are gigabytes in size. Seems like a waste of memory |
6 | Remove this functionality alltogether | The Woo guy in this article mentiones that this would be messy, as these layered navigation thingies, not only counts elements per filter, but also figures out which filters to hide or not | Seems a bad idea |
7 | Remove all filters from the shop page | Always an option, to completely stop using these product attribute filters | That's too drastic |
8 | Combine filters | We can definitely reduce the number of filters, as we have some redundancy here | This should be done anyway, but is not a definite solution |
9 | Remove some filters | We have several really big filters, and we don't know if they are all equally used | We've done this at the beginning of 2024. It seemed to have made the problem a bit less problematic. Just a bit |
10 | Reduce size of filters? | Reduce the amount of data stored in filters? That seems more relevant for the product catalog than for attribute filters, but you never know | Consider anyway, although it's not a definite solution |
11 | Disable these transients? | This post seems to suggest that you can just disable these transients. The magic keyword seems to be woocommerce_layered_nav_count_maybe_cache . This keyword is also used to check in the sourcecode before creating these transients: src » internal » ProductAttributesLookup » Filterer.ph , line 174. [10]
|
Would probably the best solution, especially if it's only a configuration setting that has to be changed |
12 | Use alternative filter plugins? | It seems there are plugins that are alternatives for the default filters on the shop page. On the other hand, they seem to have the same issues [11], [12], [13] |
Delete transients
Albeit a stop-gap measure, let's delete these transients.
wp transient delete
They may have different names at different shops, but we broadly get their names by filtering for wc_layered_nav_counts_pa_
:
wp transient list --fields=name --format=csv | grep wc_layered_nav_counts_pa_
This can be used to delete only the specific transients, and leave other transients untouched:
wp transient list --fields=name --format=csv | grep wc_layered_nav_counts_pa_ | cut -d',' -f1 | xargs -n 1 wp transient delete
Optimize database
Just removing transients through WP-CLI, is not sufficient: Without optimizing the wp_options
table, storage consumption stays the same and is likely to crash your server, no matter how big it is.
We use this:
optimize table wp_options
Script
We use a script that iterates over all webshops each hour, and executes the following code on each site:
################################################################################ # fix_dbs_and_transients ################################################################################ # fix_dbs_and_transients() { # # cd ######################################## # site_path="/var/www/${site_array[$i,url]}" echo "site_path: $site_path" cd "$site_path" # Remove transients ######################################## # echo "" echo "Transients - Before" wp transient list --fields=name echo "" echo "Transients - Deleting..." wp transient list --fields=name --format=csv | grep wc_layered_nav_counts_pa_ | cut -d',' -f1 | xargs -n 1 wp transient delete # echo "" echo "Transients - After" wp transient list --fields=name # Optimize table "wp_options" ######################################## # sql="optimize table wp_options;" echo "" echo "Optimize table wp_options - Query: $sql" echo "" wp db query "$sql" # }
This is executed through cron once per hour and the output is redirected to a log file:
52 * * * * /usr/local/bin/wp_wc_quick_fix_dbs_and_transients.sh >> /tja/example/wp_wc_quick_fix_dbs_and_transients.log 2>&1
Here is an example of the output of a storage size monitoring script (size.sh
- not the same script as mentioned above) for site at_en
:
Timestamp storage mysqldump md2 /var /var/lib /var/lib/mysql ---------------- ------- --------- --- ---- -------- -------------- 2023-12-08-11:48 125M 16.08M 36% 254G 47G 43G 2023-12-08-11:49 125M 16.33M 36% 255G 47G 43G 2023-12-08-11:50 125M 16.62M 36% 255G 47G 43G 2023-12-08-11:51 125M 16.84M 36% 255G 47G 43G 2023-12-08-11:53 125M 17.11M 36% 255G 47G 43G 2023-12-08-11:54 125M 17.36M 36% 255G 47G 43G 2023-12-08-11:55 11M 1.64M 36% 255G 47G 43G 2023-12-08-11:56 12M 1.90M 36% 255G 47G 43G 2023-12-08-11:57 15M 2.28M 36% 255G 47G 43G
The script is executed at :52 each our. Appearantly, this site is only being updated around :54, which is fine. The columns storage and mysqldump both relate to table wp_options
:
- storage:
$(sudo du -sh /var/lib/mysql/${db}/wp_options.ibd | cut -f1)
- mysqldump:
$(mysqldump ${db} wp_options | wc -c)
.
After optimizing wp_options
, storage size is about 7 times the size of the SQL-dump.
Remove transient creation
This solution is based on this posting by Roy Lindauer:
- Seems like a good solution, especially since I find it hard to believe that such a large transient is actually speeding up stuff
- Preventing these transients to be created in the first place, is much better than deleting existing transients every so often, because both creation and deletion requires CPU power
- Drawback: requires some custom coding & patching.
Patched code
It's basically out-commenting 3 lines in file src » Internal » ProductAttributesLookup » Filterer.php
around line 171:
// We have a query - let's see if cached results of this query already exist. $query_hash = md5( $query_sql ); // Maybe store a transient of the count values. $cache = apply_filters( 'woocommerce_layered_nav_count_maybe_cache', true ); if ( true === $cache ) { $cached_counts = (array) get_transient( 'wc_layered_nav_counts_' . sanitize_title( $taxonomy ) ); } else { $cached_counts = array(); } if ( ! isset( $cached_counts[ $query_hash ] ) ) { // phpcs:ignore WordPress.DB.PreparedSQL.NotPrepared $results = $wpdb->get_results( $query_sql, ARRAY_A ); $counts = array_map( 'absint', wp_list_pluck( $results, 'term_count', 'term_count_id' ) ); $cached_counts[ $query_hash ] = $counts; // // Just don't create this transients - That's all // Strompf - 2024.06 // // if ( true === $cache ) { // set_transient( 'wc_layered_nav_counts_' . sanitize_title( $taxonomy ), $cached_counts, DAY_IN_SECONDS ); // } // }
Test: nl_nl
- I had to do this on a live site, since otherwise there wouldn't be enough traffic to have a realistic case
- Choosed
nl_nl
as it is on a separate server - Applied on 2024.06.10, around 13:30
- Applied the usual script to remove all transients + optimize table:
db_navtrans_trim
. Subsequent database size: 2.4G - Disabled this script in cron around 13:40. Usually, new transients appear within seconds after removing them
- No new NavTrans transients appeared so far.
See also
- awk
- Object caching (WordPress)
- Patch-management
- Transients (WordPress)
- Wp options keeps growing (WordPress)
Sources
- https://github.com/woocommerce/woocommerce/issues/17355 - Transient for Layered Nav Counts gets to big » Primary source for this topic
- https://github.com/woocommerce/woocommerce/issues/15643 - Add option to turn off term counts for WC Layered Nav widget
- https://stackoverflow.com/questions/25202051/woocommerce-layered-nav-product-filter-widget-multiple-attributes-dont-work-a - Patching WooCommerce code suggested. Doesn't seem reliable
- https://www.roylindauer.com/blog/wp-transients-must-be-used-responsibly.html - » No, this is the primary source for this topic
- https://github.com/woocommerce/woocommerce/issues/15643 - More details about this specific query
Searching